MARK H.

HOLMES

Introduction
to Differential
Equations 2e
Introduction to Differential
Equations

Mark H. Holmes
$PQZSJHIU
ª

CZ
.BSL
)
)PMNFT

"MM
SJHIUT
SFTFSWFE

1SJOUFE
JO
UIF
6OJUFE
4UBUFT
PG
"NFSJDB

*4#/



/P
QBSU
PG
UIJT
CPPL
NBZ
CF
SFQSJOUFE
JO
BOZ
NBOOFS
XJUIPVU
XSJUUFO
QFSNJTTJPO

GSPN
UIF
QVCMJTIFS


7FOUVSF
%SJWF
4VJUF

"OO
"SCPS
.*


XXXYBOFEVDPN
 Contents

Preface v

1 Introduction 1
1.1 Terminology for Differential Equations . . . . . . . . . 2
1.2 Solutions and Non-Solutions of Differential Equations . 3

2 First-Order Equations 7
2.1 Separable Equations . . . . . . . . . . . . . . . . . . . . 7
2.1.1 General Version . . . . . . . . . . . . . . . 8
2.2 Integrating Factor . . . . . . . . . . . . . . . . . . . . . 14
2.2.1 General and Particular Solutions . . . . . . 16
2.2.2 Interesting But Tangentially Useful Topics 17
2.3 Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.1 Mixing . . . . . . . . . . . . . . . . . . . . 20
2.3.2 Newton’s Second Law . . . . . . . . . . . . 23
2.3.3 Logistic Growth or Decay . . . . . . . . . . 24
2.3.4 Newton’s Law of Cooling . . . . . . . . . . 26
2.4 Steady States and Stability . . . . . . . . . . . . . . . . 33
2.4.1 General Version . . . . . . . . . . . . . . . 35
2.4.2 Sketching the Solution . . . . . . . . . . . . 36
2.4.3 Parting Comments . . . . . . . . . . . . . . 38

3 Second-Order Linear Equations 41

3.1 Initial Value Problem . . . . . . . . . . . . . . . . . . . 42
3.2 General Solution of a Homogeneous Equation . . . . . 42
3.3 Solving a Homogeneous Equation . . . . . . . . . . . . 44
3.3.1 Two Real Roots . . . . . . . . . . . . . . . 45
3.3.2 One Real Root and Reduction of Order . . 45
3.4 Complex Roots . . . . . . . . . . . . . . . . . . . . . . 46
3.4.1 Euler’s Formula and its Consequences . . . 46
3.4.2 Second Representation . . . . . . . . . . . . 48
3.4.3 Third Representation . . . . . . . . . . . . 48
3.5 Summary for Solving a Homogeneous Equation . . . . 48
3.6 Solution of an Inhomogeneous Equation . . . . . . . . . 52

i
ii Contents

3.6.1 Non-Uniqueness of a Particular Solution . . 52

3.7 The Method of Undetermined Coefficients . . . . . . . 53
3.7.1 Finding the Coefficients . . . . . . . . . . . 56
3.7.2 Odds and Ends . . . . . . . . . . . . . . . . 56
3.8 Solving an Inhomogeneous Equation . . . . . . . . . . . 57
3.9 Variation of Parameters . . . . . . . . . . . . . . . . . . 60
3.9.1 The Solution of an IVP . . . . . . . . . . . 62
3.10 Linear Oscillator . . . . . . . . . . . . . . . . . . . . . . 64
3.10.1 The Spring Constant . . . . . . . . . . . . . 64
3.10.2 Simple Harmonic Motion . . . . . . . . . . 64
3.10.3 Damping . . . . . . . . . . . . . . . . . . . 67
3.10.4 Resonance . . . . . . . . . . . . . . . . . . 70
3.11 Euler Equation . . . . . . . . . . . . . . . . . . . . . . 76
3.11.1 Examples . . . . . . . . . . . . . . . . . . . 77
3.12 Guessing the Title of the Next Chapter . . . . . . . . . 78

4 Linear Systems 79
4.1 Linear Systems . . . . . . . . . . . . . . . . . . . . . . 80
4.1.1 Example: Transforming to System Form . . 81
4.1.2 General Version . . . . . . . . . . . . . . . 82
4.2 General Solution of a Homogeneous Equation . . . . . 84
4.3 Review of Eigenvalue Problems . . . . . . . . . . . . . 85
4.4 Solving a Homogeneous Equation . . . . . . . . . . . . 91
4.4.1 Complex-Valued Eigenvalues . . . . . . . . 92
4.4.2 Defective Matrix . . . . . . . . . . . . . . . 93
4.5 Summary for Solving a Homogeneous Equation . . . . 94
4.6 Phase Plane . . . . . . . . . . . . . . . . . . . . . . . . 99
4.6.1 Examples . . . . . . . . . . . . . . . . . . . 99
4.6.2 Connection with an IVP . . . . . . . . . . . 104
4.7 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . 107

5 Nonlinear Systems 111

5.1 Non-Linear Systems . . . . . . . . . . . . . . . . . . . . 112
5.1.1 Steady-State Solutions . . . . . . . . . . . . 114
5.2 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . 116
5.2.1 Derivation of the Stability Conditions . . . 117
5.2.2 Summary . . . . . . . . . . . . . . . . . . . 121
5.2.3 Examples . . . . . . . . . . . . . . . . . . . 122
5.3 Periodic Solutions . . . . . . . . . . . . . . . . . . . . . 128
5.3.1 Closed Solution Curves and Hamiltonians . 131
5.3.2 Finding the Period . . . . . . . . . . . . . . 135
5.4 Motion in a Central Force Field . . . . . . . . . . . . . 139
5.4.1 Steady States . . . . . . . . . . . . . . . . . 141
5.4.2 Periodic Orbit . . . . . . . . . . . . . . . . 141
Contents iii

6 Laplace Transform 145

6.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . 145
6.1.1 Requirements . . . . . . . . . . . . . . . . . 146
6.1.2 Examples . . . . . . . . . . . . . . . . . . . 147
6.2 Inverse Laplace Transform . . . . . . . . . . . . . . . . 150
6.2.1 Jump Discontinuities . . . . . . . . . . . . 153
6.2.2 Heaviside Expansion Theorem . . . . . . . 155
6.3 Properties of the Laplace Transform . . . . . . . . . . . 157
6.3.1 Transformation of Derivatives . . . . . . . . 157
6.3.2 Convolution Theorem . . . . . . . . . . . . 158
6.4 Solving Differential Equations . . . . . . . . . . . . . . 159
6.4.1 The Transfer Function . . . . . . . . . . . . 161
6.4.2 Comments and Limitations on Using the
Laplace Transform . . . . . . . . . . . . . . 162
6.5 Solving Equations with Non-Smooth Forcing . . . . . . 164
6.5.1 Impulse Forcing . . . . . . . . . . . . . . . 165
6.6 Solving Linear Systems . . . . . . . . . . . . . . . . . . 170
6.6.1 Chapter 4 versus Chapter 6 . . . . . . . . . 172

7 Partial Differential Equations 175

7.1 Balance Laws . . . . . . . . . . . . . . . . . . . . . . . 176
7.2 Boundary Value Problems . . . . . . . . . . . . . . . . 176
7.2.1 Eigenvalue Problems . . . . . . . . . . . . . 178
7.3 Separation of Variables . . . . . . . . . . . . . . . . . . 180
7.3.1 Separation of Variables Assumption . . . . 181
7.3.2 Finding F (x) and λ . . . . . . . . . . . . . 182
7.3.3 Finding G(t) . . . . . . . . . . . . . . . . . 183
7.3.4 The General Solution . . . . . . . . . . . . 183
7.3.5 Satisfying the Initial Condition . . . . . . . 183
7.3.6 Examples . . . . . . . . . . . . . . . . . . . 184
7.4 Sine and Cosine Series . . . . . . . . . . . . . . . . . . 188
7.4.1 Finding the bn ’s . . . . . . . . . . . . . . . 188
7.4.2 Convergence Theorem . . . . . . . . . . . . 189
7.4.3 Examples . . . . . . . . . . . . . . . . . . . 189
7.4.4 Cosine Series . . . . . . . . . . . . . . . . . 193
7.4.5 Differentiability . . . . . . . . . . . . . . . . 196
7.4.6 Infinite Dimensional . . . . . . . . . . . . . 197
7.5 Wave Equation . . . . . . . . . . . . . . . . . . . . . . 199
7.5.1 Examples . . . . . . . . . . . . . . . . . . . 202
7.5.2 Natural Modes and Standing Waves . . . . 203
7.6 Inhomogeneous Boundary Conditions . . . . . . . . . . 204
7.6.1 Steady State Solution . . . . . . . . . . . . 205
7.6.2 Transformed Problem . . . . . . . . . . . . 205
7.6.3 Summary . . . . . . . . . . . . . . . . . . . 206
iv Contents

7.6.4 Wave Equation . . . . . . . . . . . . . . . . 207

7.7 Inhomogeneous PDEs . . . . . . . . . . . . . . . . . . . 208
7.7.1 Summary . . . . . . . . . . . . . . . . . . . 209
7.7.2 A Very Useful Observation . . . . . . . . . 211
7.8 Laplace’s Equation . . . . . . . . . . . . . . . . . . . . 212
7.8.1 Rectangular Domain . . . . . . . . . . . . . 213
7.8.2 Circular Domain . . . . . . . . . . . . . . . 217

A Matrix Algebra: Summary 223

B Answers 225

Bibliography 236

Index 237
 Preface

This textbook is written for an introductory, or beginning, course in

differential equations. It is more concise than most textbooks at this level.
The reason is that most books are encyclopedic, and this enables them
to also be used in more advanced courses. The approach taken here is to
concentrate on the intended audience, and leave the additional material to
textbooks written explicitly for the more advanced, or specialized, courses
(and there are some very good ones available). It should also be pointed
out that the greater length means that they are more expensive.
One of the principal objectives of the text is fairly simple, and it
is, given a differential equation, find the solution. Most of the text is
dedicated to doing exactly this. The reason there is more than one chapter
is that the way you solve the equation depends on what type of equation
it is. A second important objective is to be able to determine the basic
geometrical properties of the solution, and to be able to do this from the
differential equation. This is important because real-world problems often
involve equations that are difficult to solve, or you can solve them but the
formula for the solution is complicated. In such cases, to be able to infer
the basic properties of the solution directly from the differential equation
is invaluable. If you want an example of what this means, read Section
2.4.
There are students, even very good ones, who do not read a lot of
what is written in a mathematics textbook. If you are one of them, here
are some tips. First, any text that is in bold font, make sure you read
what it says. As a second tip, the table below is a listing of the 10 most
often used words or phrases in the text related to differential equations.
Given that there are only about 220 pages of text, these words are used
a lot. Make sure, when the word or phrase is first used, that you know
what it means.
The prerequisites for this text vary with the chapter. The basic re-
quirement is calculus, and it is essential that this includes integration
rules such as integration by parts and partial fractions. The material re-
quiring the calculus of vector-valued functions is in Chapters 4 and 5, and
at the end of Chapter 6. These chapters also require an understanding
of a few of the elementary properties of matrices, and a short summary

v
vi Preface

Word or Expression Used

solution(s) 805
differential equation(s) 232
linear/nonlinear 223
steady state 197
stable/unstable/stability 180
eigenvalue(s) 161
general solution 155
initial value problem/IVP 105
initial condition(s) 104
homogeneous/inhomogeneous 96
Table 1. Number of times the word, or phrase, is used in this textbook.

of what you need to know is given in Appendix A. It is not necessary to

have taken a course in matrix algebra. However, there is a fundamental
connection between differential equations and linear algebra, and this con-
nection is used throughout this textbook. The material is written so it is
self-contained, so a previous course in linear algebra is not necessary. You
will see comments, such as “if you recall from linear algebra,” which are
used to indicate where the connections are, but the material required for
differential equations is then written out explicitly. Occasionally there
are facts, or results, from linear algebra that are needed and they are
stated without proof. This is also done with other topics, and in such
cases references are often given where you can find out more about the
subject.
A computer, or computer software, are not required anywhere in this
text. There are, however, a small number of exercises that require you to
evaluate a mathematical expression using a calculator.
There is a web-page for the text, and it is reachable via the author’s
web-page. It includes plots needed for some of the exercises, videos for
Chapter 7, and, assuming there are any, a listing of the typos.
I would like to thank Peter Kramer for numerous, very useful, sugges-
tions on how to improve the text. Also, as usual, I would like to thank
those who developed and have maintained TeXShop, a free and very good
TeX previewer.

Mark H. Holmes
Department of Mathematical Sciences
Rensselaer Polytechnic Institute
May, 2020
Chapter 1

Introduction

We begin with a question: why are most students who are majoring
in engineering or science required to take an entire course dedicated to
something called differential equations?

We’ll start to answer this by giving a couple of examples where they

arise, and this will also provide an opportunity to introduce some of the
terminology used in the subject.

Example 1: Rate Laws

These describe the fluctuations, or changes, in something. The something
in this case could be the concentration of a chemical, a population of
animals, or perhaps the temperature of an object. As a simple example,
a radioactive isotope is unstable, and will decay by emitting a particle,
transforming into another isotope. The assumption usually used to model
such situations is that the rate of decrease in the amount of radioactive
isotope is proportional to the amount currently present. To translate this
into mathematical terms, let N (t) designate the amount of the radioactive
material present at time t. In this case we obtain the rate equation

dN
= −kN, for 0 < t, (1.1)
dt
where k is a positive constant. This is a differential equation for N .
Usually one knows the amount N0 of the isotope at the beginning, which
gives us the requirement that

N (0) = N0 . (1.2)
Introduction to Differential Equations, M. H. Holmes, 2020

1
2 Chapter 1. Introduction

This is known as an initial condition. Together, (1.1) and (1.2) form

what is called an initial value problem (IVP). 

Example 2: Mechanics
One of the biggest generators of differential equations is Newton’s second
law, which states that F = ma. Any situation, electrical, mechanical or
otherwise, involving non-static forces will almost inevitability result in
having to solve a differential equation. To illustrate, consider the simple
case of dropping an object off a building. If x(t) is the distance of the
object from the ground, then its velocity is v = x′ (t), and its acceleration
is a = x′′ (t). If the forces on the object are gravity Fg = −mg, and air
resistance Fr = −cv, then F = Fg +Fr . Together, these expressions result
in the following differential equation for x(t):

d2 x dx
m 2
= −mg − c .  (1.3)
dt dt

The differential equations in (1.1) and (1.3) have a few things in com-
mon, such as there is one independent variable, t, and one dependent vari-
able, N and x. There are also differences, and an example is that (1.3)
involves the second derivative and (1.1) only involves the first derivative.
It is important to be able to recognize these differences as they are often
used in this textbook to determine how to solve the problem.

1.1 Terminology for Differential Equations

Problems involving differential equations can involve a single equation,
or several equations. They can also have one independent variable, or
several. There are other differences, and to help illustrate some of the
possibilities we will use the following examples.

d2 y dy ∂u ∂u
Example 1: 2
− 2 + 4ty = 0 Example 2: −2 = u3
dt dt ∂t ∂x
du
Example 3: =u+v+1
dt
dv
= −u + v
dt

Dependent variable(s): This is the variable(s) being solved for.

Example 1: y Example 2: u Example 3: u and v

1.2. Solutions and Non-Solutions of Differential Equations 3

Independent variable(s): These are usually time (t) and/or space (x).

Example 1: t Example 2: x and t Example 3: t

Order: The order of the highest derivative in the equation (or equations).

Example 1: second-order Example 2: first-order

Example 3: first-order

Linear or Nonlinear: A differential equation is linear if it is a linear

expression of the dependent variable and its derivatives, otherwise
it is nonlinear.

Example 1: linear Example 2: nonlinear (because of the u3 )

Example 3: linear

ODE or PDE: If there is one independent variable, then it is an ordi-

nary differential equation (ODE). If there is more than one inde-
pendent variable, then it is a partial differential equation (PDE).

Example 1: ODE Example 2: PDE Example 3: ODEs

Homogeneous or Inhomogeneous: A linear differential equation is

homogeneous if the identically zero function is a solution. Oth-
erwise, it is inhomogeneous.

Example 1: homogeneous since y ≡ 0 is a solution

Example 2: inapplicable since the equation is not linear
Example 3: inhomogeneous since u ≡ 0 and v ≡ 0 is not a solution

1.2 Solutions and Non-Solutions of Differential Equations

One of the central questions of this textbook is how to find the solution
of a differential equation. The examples below are about the reverse
situation, where a function is given and the question is whether it is a
solution of a particular differential equation.

Example 1: Show that y = te−2t is a solution of y ′ = −2y + e−2t .

Answer: Since
y ′ = e−2t − 2te−2t = (1 − 2t)e−2t ,
−2y + e−2t = −2te−2t + e−2t = (1 − 2t)e−2t ,
it follows that y ′ = −2y + e−2t (i.e., y is a solution). 
4 Chapter 1. Introduction

Example 2: For what value(s) of r and c, if any, is y = cert a solution

of the IVP: y ′ + y = 0, where y(0) = 3?
Answer: Since y ′ = rcert , then from the differential equation we
require that rcert + cert = 0. This can be written as (r + 1)cert = 0.
Given that ert is never zero, we conclude that either c = 0 or else
r = −1. From the initial condition y(0) = 3, we need c = 3, and so
this means that r = −1. 

Example 3: For what value(s) of r, if any, is y = ert a solution of the

equation y ′′ − y ′ − 6y = 0?
Answer: Since y ′ = rert , and y ′′ = r2 ert , then from the differential
equation we require that (r2 − r − 6)ert = 0. Given that ert is never
zero, we conclude that r2 −r −6 = 0. Solving this, we get that r = 3
and r = −2 are the only values for which y = ert is a solution. 

Example 4: For what value(s) of r and c, if any, is y = ert a solution of

y ′ = 2y 3 ?
Answer: Since y ′ = rert , then from the differential equation we
require that rert = 2e3rt . Given that ert is never zero, we need
r = 2e2rt . The left hand side is constant. The only way to have
the right hand side a constant is to take r = 0. In this case, the
differential equation becomes 0 = 2. This is not possible, and so the
answer is that no values result in a solution. 

Exercises
1. Show that the given function y(t) is a solution of the given differential
equation.

a) y = e2t − 1, y ′ = 2y + 2 e) y = et + 1, y ′′ + 2y ′ − 3y = −3
1
b) y = te−t , y ′ + y = e−t f) y = , y′ + y2 = 0
1+t
c) y = cos(3t), y ′′ = −9y g) y = tan 31 t + 1 , 3y ′ = 1 + y 2

d) y = e3t , y ′′ + y ′ − 12y = 0 h) y = ln(1 + t2 ) , y ′ = 2te−y

2. For what value(s) of r, if any, is y = ert a solution of the differential

equation?

a) y ′ = −2y f) y ′′ − 4y ′ + 4y = 0
b) 3y ′ = y g) y ′′ + y ′ + y = e−3t
c) y′ = y + 1 h) y ′′ − 3y ′ + y = 1
d) y ′′ + 4y ′ = 0 i) y ′ = −2y 3
e) 2y ′′ + 5y ′ − 3y = 0 j) y′ = y2
Exercises 5

3. For what value of r and c is y = cert a solution of the IVP?

a) y ′ = −2y, y(0) = 1 d) y ′ − y = 0, y(0) = −1

b) y ′ + y = 0, y(0) = −1 e) 5y ′ = −2y, y(0) = −7
c) 3y ′ − y = 0, y(0) = 3 f) y′ + 4y = 0, y(0) = 3

4. The following are linear and homogeneous first-order differential equa-

tions. The given function y1 (t) is a solution, and you are to show that
y = cy1 is a solution for any value of the constant c.

a) y ′ = 2y, y1 = e2t c) y ′ − 4y = 0, y1 = e4t

b) y ′ + y = 0, y1 = e−t d) 3y ′ = y, y1 = et/3

5. The following are linear and homogeneous second-order differential

equations. The given functions y1 (t) and y2 (t) are solutions, and you
are to show that y = c1 y1 + c2 y2 is a solution for any value of the
constants c1 and c2 .

a) y ′′ − 3y ′ + 2y = 0, c) y ′′ + y ′ = 0,
y1 = e2t , y2 = et y1 = e−t , y2 = 1
b) y ′′ − y ′ − 2y = 0, d) y ′′ + 2y ′ + 5y = 0,
y1 = e2t , y2 = e−t y1 = e−t cos(2t)
y2 = e−t sin(2t)

Important Conclusion: Problems 4 and 5 are demonstrations of the

fact that if y1 (t) and y2 (t) are solutions of a linear and homogeneous
differential equation, then c1 y1 (t)+ c2 y2 (t) is a solution of the equation
for any value of c1 and c2 . This is known as the principle of super-
position, and it holds for all linear homogeneous differential equations
(ODEs or PDEs). Moreover, as demonstrated in the following exercise,
this does not (usually) hold for a nonlinear differential equation.

6. Both y1 (t) and y2 (t) are solutions of the given nonlinear differential
equation. Show that (i) y = c1 y1 (t) is not a solution unless c1 = 1,
and (ii) y = c1 y1 + c2 y2 is not a solution if c1 and c2 are both nonzero.
√
a) y ′ = t/(1 + y), b) y ′ = 1 + y,
y1 = −1 + t, y2 = −1 − t y1 = 14 t2 + t, y2 = 41 t2 + 2t + 3

7. Fill out the table on the next page. Assume that any constants in the
equation(s) are nonzero. Also, in the last column, the answer Inap-
plicable (IA) is possible. Reference for Schrödinger’s equation image:
Eigler [2020].
 6

Equation(s) dep indep order linear (L) or ODE or homog (H) or

var(s) var(s) nonlinear (NL) PDE inhomog (IH)

Radioactive decay
dy
dt
= −ry

Mass-Spring-Dashpot system
d2 y dy
m 2 + c + ky = sin t
dt dt

Pendulum equation
d2 θ g
dt2 ℓ
= − sin θ

Schrödinger’s equation
∂u ~2 ∂ 2 u
i~ +Vu
∂t 2m ∂x2
=−

Beam equation
∂4w ∂2w
+ 2 =P
∂x4 ∂t

Michaelis-Menten equations
dS
dt
= −k1 ES + k−1 (E0 − E)

dE
Chapter 1. Introduction

dt
= −k1 ES + k3 (E0 − E)
 Chapter 2

First-Order Equations

This chapter concerns solving differential equations of the form

dy
= f (t, y).
dt
There are no known analytical methods that can solve the general ver-
sion of this problem. Consequently, assumptions have to be made on
the function f (t, y) to be able to derive a solution. The two more useful
assumptions are that the function is separable or it is linear, and both
are considered in this chapter. The fact is, however, that for many real
world problems it is not possible to solve the differential equation by
hand. Consequently, the ability to determine the properties of the solu-
tion, without actually solving the problem, becomes essential. What this
entails is introduced in Section 2.4.

2.1 Separable Equations

To introduce this method we begin by considering the differential equation

dy
= 3y 2 . (2.1)
dt
We are going to treat the derivative as if it were a fraction, and rewrite
the above equation as
dy
= 3dt. (2.2)
y2
So, the variables have been separated in the sense that all of the y terms
are on the left hand side, and the t terms are on the right. We now
Introduction to Differential Equations, M. H. Holmes, 2020

7
8 Chapter 2. First-Order Equations

integrate both sides, which gives

dy
Z Z
= 3dt.
y2
Carrying out the integrations, and including the usual integration con-
stant, we have
1
− = 3t + c. (2.3)
y
Solving this for y, we obtain the solution
1
y=− . (2.4)
3t + c
The last step is to check on whether the separation of variables step might
involve dividing by zero. This happens for (2.2) when y = 0. Moreover,
the constant function y = 0 is a solution of (2.1), and it is not included
in (2.4). Consequently, another solution of the differential equation is

y = 0. (2.5)

The method used to solve (2.1) is rather simple, but it contains the
questionable step of splitting the derivative to obtain (2.2). To explain
why this is possible, note that (2.1) can be written as y −2 dy
dt = 3. Using
d
the chain rule, this can be written as − dt (y −1 ) = 3. Integrating this
equation yields (2.3). So, the splitting the derivative step is effectively a
compact version of using the chain rule.

2.1.1 General Version

To explain how the method can be used for other problems, suppose the
differential equation to solve is
dy
= f (t, y). (2.6)
dt
The method requires that it is possible to find a factorization of the form
f (t, y) = F (t)G(y). This means that it is possible to write the differential
equation as
dy
= F (t)G(y). (2.7)
dt
Separating variables gives
dy
= F (t)dt,
G(y)
and integrating we get
dy
Z Z
= F (t)dt. (2.8)
G(y)
2.1. Separable Equations 9

In theory, you carry out the above integrations, and then solve for y.
How difficult this might be depends on how complicated the y integral
is, and the examples that follow illustrate some of the complications that
can arise. It is also important to note that the above method requires
that G(y) 6= 0. Consequently, in addition to the solutions that come from
(2.8), you must include as solutions any constant that satisfies G(y) = 0.

Example 1: Find all solutions of 4y ′ = −y 3 .

Answer: Since f (t, y) = − 14 y 3 , we can take F (t) = 1
4 and G(y) =
−y 3 . So, (2.8) becomes

dy 1
Z Z
− = dt.
y3 4

Integrating gives us
1 1
2
= t + c,
2y 4
which is rewritten as
2
y2 = .
t + 4c
From this we obtain the two solutions
r
2
y=± . (2.9)
t + 4c

To check on the G(y) = 0 solutions, solving G(y) = 0 gives y = 0.

This constant function is not included in the above expressions for
y, so it is a third solution of the equation. 

Example 2: Find the solution of the IVP: 4y ′ = −y 3 , where y(0) = −3.

Answer: The three solutions of the differential equation were de-
rived in the previous example. Because the initial condition requires
the solution to be negative, the solution we need is
r
2
y=− .
t + 4c
√
Setting y = −3 and t = 0 in this equation gives 3 = 1/ 2c, which
means that c = 1/18. Therefore, the solution is
r
18
y=− . 
9t + 2
10 Chapter 2. First-Order Equations

Example 3: Is y ′ + y = t a separable equation?

Answer: No. For this equation, f (t, y) = t − y, and it is not possible
to factor this as f (t, y) = F (t)G(y). How to solve this equation is
explained in the next section. 

dw x
Example 4: Solve = , where w(0) = −2.
dx 1+w
Answer: In this problem the independent variable is x and the
dependent variable is w. Separating variables, so (1 + w)dw = xdx,
and then integrating gives
Z Z
(1 + w)dw = xdx.

Carrying out the integrations we get that

1 1
w + w2 = x2 + c.
2 2
To satisfy the initial condition, substitute w = −2 and x = 0 into
the above equation, from which we get that c = 0. This leaves
w + 12 w2 = 21 x2 , or equivalently, w2 + 2w − x2 = 0. This is a
quadratic equation in w, and solving it we get the two solutions
p
w = −1 ± 1 + x2 .

The initial condition is needed to determine which sign to use, and

since w(0) = −2 then we need √ the minus sign. Therefore, the solu-
tion of the IVP is w = −1 − 1 + x2 . 

y
Example 5: Solve y ′ = − , where y(0) = 1.
1+y
Answer: Separating variables yields
1+y
dy = −dt.
y
Since (1 + y)/y = 1/y + 1, and y(0) > 0, then integrating we get
that
y + ln y = −t + c.
It is not possible to solve this for y as in the previous examples,
without resorting to more advanced mathematical methods. For
this reason, this is an example of what is called an implicit solu-
tion, and they are very common when solving nonlinear differential
equations. Even so, it is still possible to find c from the initial con-
dition. Substituting y = 1 and t = 0 into the above equation we get
2.1. Separable Equations 11

that c = 1. Therefore, the solution of the IVP is defined implicitly

through the equation

y + ln y = −t + 1.  (2.10)

A few comments need to be made about separation of variables before

ending this section.

Integration Constant: The integration constant plays an essential role

in the solution of a differential equation. It is useful to be aware that
there are different ways you can write it. As an example, instead of
(2.9), you can write the solution as
r
2
y=± ,
t + c̄

where c̄ = 4c. Similarly, if the solution is found to be

3t − 2c + 4
y= ,
t + 2c − 4

you can write it as

3t − c̄
y= , (2.11)
t + c̄
where c̄ = 2c − 4. For both of these examples, the solution contains
one undetermined constant, just as in the original version of each
solution. It should also be mentioned that this simplification is often
used when giving the answers to the exercises. Moreover, instead of
(2.11), the answer will likely be written as

3t − c
y= .
t+c

Linear or Nonlinear: The method works on linear and nonlinear first-

order differential equations. However, it does not work on every
linear or nonlinear equation.

Non-uniqueness of Factorization: The factorization f (t, y) = F (t)G(y)

is not unique. For example, for f (t, y) = y + ty you can take
F (t) = 1 + t and G(y) = y. You can also take F (t) = 12 (1 + t)
and G(y) = 2y. It makes no difference which one you use, it is just
required that f (t, y) = F (t)G(y). Any such factorization will lead,
eventually, to the same, or an equivalent, solution of the differential
equation.
12 Chapter 2. First-Order Equations

Existence and Uniqueness: When solving an IVP there is always the

question of whether there is a solution (existence), and whether
there is more than one solution (uniqueness). It is possible to find
problems that have no solution (see Exercise 6(b)), have a solution
only for 0 ≤ t < T (see Exercise 6(c)), or have multiple solutions
(see Exercise 6(d)). It is important to be aware of this, but the
theory underlying existence and uniqueness is beyond the purview
of this text. It is, however, a topic covered in most upper division
courses on ODEs.

Exercises
1. Find all of the solutions of the given differential equation.

a) y ′ = −3y 4 f) (1 + t)y ′ = −e3y k) 2y ′ = y 2 − 6y + 9

b) y ′ = y 3 e−t g) y ′ = −e2t+4y l) 3y ′ = y 2 + 1
c) y ′ + y 2 sin t = 0 h) y ′ = −2y m) y ′ − tey = te−y
d) 2y ′ = t/(y − 3) i) y ′ + (1 + 3y)3 = 0 n) y ′ − e−y = 1
e) y ′ = −(2 + t)ey j) y ′ = y 2 + 4y + 4 o) y ′ = t(y + 1/y)

2. Find the solution of the IVP.

a) y ′ = −y 3 , y(0) = 5 g) y ′ = 1 + cos(y), y(0) = π/2

b) y ′ = −2y 3 , y(0) = 0 h) y ′ = y 2 − 5y, y(0) = 1
c) (1 + t)y ′ = 3 + y, y(0) = 4
i) y ′ + e−2y = 1, y(0) = 1
d) (4 + et )y ′ + et y 2 = 0, y(0) = 1
j) y ′ = 1/(e−y + ey ), y(0) = 0
e) y′ = te−y , y(0) = −1 p
1 k) y ′ = 1 − y 2 , y(0) = 0
f) y ′ = , y(0) = 0 Hint: y ′ ≥ 0
2+y

3. Find the solution of the IVP. In these problems, the independent vari-
able is not t and the dependent variable is not y.

dq dz
a) = −7q 3 , q(0) = −1 e) (1+e−r ) +z 2 = 0, z(0) = 6
dr dr
dp dw
b) = −4p3 , p(0) = 0 f) 4 = τ 3 e−2w , w(0) = 0
dr dτ
dh dr
c) 3 = 2 + h, h(0) = 2 g) (θ + 1)3 = r2 , r(0) = 2
dτ dθ
dh dr 2θ
d) = h2 − 3h, h(0) = 2 h) = , r(0) = 0
dx dθ 1+r
Exercises 13

4. Find the solution of the IVP in implicit form.

1 1+y
a) y ′ = 1 + , y(0) = 1 c) y ′ = , y(0) = 5
y 2+y
3 ey
b) y ′ = , y(0) = −1 d) y ′ = , y(0) = 2
1 + y4 1 + ey
y-axis

x-axis

Figure 2.1. Cable hanging between two poles, as described in Exercise 5.

5. A cable is hung between two poles as illustrated in Figure 2.1. The

poles are located at x = −L and x = L, and each has height h. The
curve y(x) determined by the cable minimizes the cable’s potential
energy. From this, one obtains the equation
r
d2 y  dy 2
a 2 = 1+ , for − L < x < L,
dx dx
where a is a positive constant. Because of the symmetry in the prob-
lem, y ′ (0) = 0.
a) Letting w(x) = y ′ (x), rewrite the differential equation as a first-
order equation involving w and w′ . Also, what is w(0)?
b) Solve the problem in part (a) for w.
c) Integrate y ′ (x) = w(x), and use the condition y(L) = h, to deter-
mine y(x). The solution you are finding is an example of what is
called a catenary.
6. The following illustrate some of the complications that can arise when
solving nonlinear differential equations.
a) The question is whether the implicit solution (2.10) actually has a
solution. To show this, rewrite (2.10) as ln y = −y − t + 1. Setting
g(y) = ln y, and h(y) = −y − t + 1, let t = 0 and then sketch g(y)
and h(y) on the same axes for 0 < y < ∞. Explain why this shows
that there is exactly one solution of (2.10). Do the same thing for
t = 1 and t = 2. Use this sketching procedure to determine what
value y approaches as t → ∞.
14 Chapter 2. First-Order Equations

b) Consider the IVP: ty ′ = y + 1, where y(0) = 1. Try solving this and

show that there is no solution.
c) Solve y ′ = 21 y 3 , where y(0) = 1. Explain why there is no solution
for t ≥ 1.
d) Show that there are an infinite number of solutions of ty ′ = y + 1,
where y(0) = −1.

2.2 Integrating Factor

The equation to be solved is

y ′ + p(t)y = g(t). (2.12)

What is important here is that this equation is linear, as well as first-order.

The solution will be derived using two formulas from calculus. The
first is the product rule, which states that

d
(µy) = µ(t)y ′ (t) + µ′ (t)y(t). (2.13)
dt
The second is the Fundamental Theorem of Calculus, which states that if

d
(µy) = q(t),
dt
then Z t
µy = q(s)ds + c. (2.14)
0

The first step is the observation that the left hand side of (2.12) re-
sembles the right hand side of (2.13). To make it so they are exactly the
same we need to multiply the differential equation by µ(t), which gives us

µy ′ + µpy = µg. (2.15)

What we need, to get this to work, is that µ must be such that

µ′ = pµ. (2.16)

It will make the formula for the solution a bit simpler if we require

µ(0) = 1. (2.17)

The differential equation (2.16) is separable, and one finds that the solu-
tion that satisfies (2.17) is
Rt
p(r)dr
µ(t) = e 0 . (2.18)
2.2. Integrating Factor 15

With this choice for µ, the differential equation for y in (2.15) can be
written as
d
(µy) = µg. (2.19)
dt
From (2.14) we get that
Z t
µy = µ(s)g(s)ds + c,
0

where c is the usual integration constant. The solution of (2.12) is there-

fore
Z t 
1
y(t) = µ(s)g(s)ds + c . (2.20)
µ(t) 0

The function µ(t), which is given in (2.18), is said to be an integrating

factor for the original differential equation.
There are two important special cases to mention. First, suppose that
the problem has an initial condition, say y(0) = y0 . Since µ(0) = 1, then
from (2.20) the solution of the resulting IVP is
Z t 
1
y(t) = µ(s)g(s)ds + y0 . (2.21)
µ(t) 0

The second special case arises for the homogeneous equation y ′ +p(t)y = 0.
Setting g = 0 in (2.20), gives us the solution
Rt
p(r)dr
y(t) = ce− 0 . (2.22)

If y(0) = y0 , then the resulting solution is

Rt
p(r)dr
y(t) = y0 e− 0 . (2.23)

Example 1: Solve y ′ + 3y = e2t .

Answer: Since p = 3, then
Z t Z t
p(r)dr = 3dr = 3t.
0 0

From (2.18), the integrating factor is µ = e3t . So, since g(t) = e2t ,
then from (2.20),
Z t  Z t 
−3t 3s 2s −3t 5s
y(t) = e e e ds + c = e e ds + c .
0 0
16 Chapter 2. First-Order Equations

Carrying out the integration,

   
−3t 1 5s −3t 1 5t 1
t
y(t) = e e +c =e e − +c
5 s=0 5 5
1
= e2t + c̄e−3t ,
5
where c̄ = c − 1/5 is an arbitrary constant. 

Example 2: Solve 2y ′ − ty = 6, where y(0) = 5.

2 /4
Answer: Since p = −t/2, then from (2.18), µ = e−t . Given that
g = 3, then from (2.21) we have
Z t 
2 /4 2 /4
y(t) = et 3e−s ds + 5 .
0

The integral in the above expression can not be written in terms of

elementary functions, and so that is the final answer. 

dh
Example 3: Solve − 4h = 2z, where h(0) = −1.
dz
Answer: In this problem the independent variable is z and the
dependent variable is h. The formula for the solution can still be
used, we just need to make the appropriate substitutions. Since
p = −4, then Z z Z z
p(r)dr = −4dr = −4z.
0 0

From (2.18), the integrating factor is µ = e−4z . So, since g(z) = 2z,
then from (2.21),
Z z 
4z −4s
h(z) = e 2se ds − 1
0
1 7
= − (4z + 1) − e4z . 
8 8

2.2.1 General and Particular Solutions

We have shown that the solution of the linear differential equation

y ′ + p(t)y = g(t), (2.24)

is Z t 
1
y(t) = µ(s)g(s)ds + c . (2.25)
µ(t) 0
2.2. Integrating Factor 17

Any, and all, solutions of (2.24) are included in this formula, and for this
reason (2.25) is said to be the general solution.
A useful observation about (2.25) is that it can be written as

y(t) = yp (t) + yh (t), (2.26)

where
t
1
Z
yp (t) = µ(s)g(s)ds, (2.27)
µ(t) 0
and Rt
c
yh (t) = = ce− 0 p(r)dr . (2.28)
µ(t)
The formulas for yp and yh are not important here. What is important is
that yp is a solution of the differential equation (2.24). It does not contain
the arbitrary constant, and for this reason it is said to be a particular
solution. In contrast, the function yh (t), which contains an arbitrary
constant, is a solution of the differential equation

y ′ + p(t)y = 0. (2.29)

This is the homogeneous equation coming from (2.24). Consequently,

yh (t) is said to be the general solution of the associated homoge-
neous equation.

Example 4: In Example 1 we found that the general solution is

1
y(t) = e2t + c̄e−3t ,
5
where c̄ is an arbitrary constant. In this case, a particular solution
is yp = 15 e2t , and the general solution of the associated homogeneous
equation is yh = c̄e−3t . 

The observation in the previous paragraph that the general solution

can be written as the sum of a particular solution and the general solution
of the associated homogeneous equation holds for all linear differential
equations (not just those that are first-order). Because we are able to de-
rive a formula for the solution, which is given in (2.25), this observation
is not really needed to solve first-order linear differential equations. How-
ever, for second-order equations, which will be studied in the next chapter,
this observation serves a fundamental role in finding the solution.

2.2.2 Interesting But Tangentially Useful Topics

The following topics are worth knowing about. However, you can skip this
material, if you wish, as it is not required to solve any of the problems in
this chapter.
18 Chapter 2. First-Order Equations

Method of Undetermined Coefficients

Most first-order linear differential equations that arise in applications have

constant coefficients, which means that they can be written as

y ′ + ay = g(t), (2.30)

where a is a constant. This can be solved using an integrating factor, but

there is often an easier way to find the solution. This involves making
an educated guess for the particular solution. The guess depends on the
specific form of the function g(t), and it is the basis of what is called the
method of undetermined coefficients. This is explained in Section 3.7 for
second-order equations, but it works on first-order equations as well. The
reason it is easier is that it avoids having to integrate anything, and you
therefore do not need to remember integration rules to find the solution.
In fact, for a problem such as the one in Example 1, you can solve it in
your head and simply write the answer down. On the other hand, the
method will not work for Example 2, and √ it will not work for Example 1
2t
if g(t) = e is replaced, say, with g(t) = t. If you want to pursue this
idea a bit more, after reading Section 3.7, you should look at Exercise 4
on page 59.

Connections with Linear Algebra

For those who have taken a course in linear algebra, there is a connec-
tion between that subject and linear differential equations that is worth
knowing about. To explain, a central problem in linear algebra is to solve
Ax = b, where A is a m × n matrix. It’s possible to prove that if there
is a solution of this equation, then it has the form x = xp + xh , where
xp is a particular solution and xh is the general solution of the associated
homogeneous equation Ax = 0. This is basically the same statement we
made for the solution of the linear differential equation (2.24). The key
property these equations have in common is that they are both linear. A
consequence of this is that the principle of superposition can be used (see
page 5) when solving the associated homogeneous equation. We will make
use of this fact in every chapter of this textbook, except for Chapter 5.
This illustrates the beauty, and profundity, of mathematical abstraction.
Namely, it is possible to make rather significant conclusions about the so-
lution of an equation, irrespective of whether it is algebraic or differential,
simply from the basic properties these equations have in common.

Exercises
1. Find the general solution of the given differential equation.
Exercises 19

a) y ′ + 3y = 0 e) (3t + 2)y ′ + 3y = sin(4t) + 5

b) y ′ − 2y = t f) (2 + t)y ′ + y = 1
√
c) 4y ′ − y = 6 + 2t g) y ′ − 3y = 1 + t
t
d) y ′ = −y + 2et − 1 h) 2y ′ + y = 1+t

2. Find the solution of the IVP.

a) y ′ − y = 4 , y(0) = −1 d) 2y ′ = y + e−t − 2, y(0) = 1

b) y ′ + 4y = 3t, y(0) = 0 e) (5 + t)y ′ + y = −1, y(0) = 2
c) 5y ′ + y = 0, y(0) = 2 f) 3y ′ + ty = −2, y(0) = 0

3. Find the solution of the IVP. In these problems, the independent vari-
able is not t and the dependent variable is not y.

dq dz
a) + 2q = 4 , q(0) = −1 d) = 4z + 1 + τ , z(0) = 0
dz dτ
dp dh
b) + 4p = −x, p(0) = 0 e) (x + 7) + h = −1, h(0) = 2
dx dx
dw dh
c) 2 − w = e2τ , w(0) = 0 f) (5z+1) +5h = 3, h(0) = −1
dτ dz

4. Find a particular solution, and the general solution to the associated

homogeneous equation, of the following differential equations.

a) y ′ − 2y = 6 c) 7y ′ − y = e2t + 3
b) y ′ + y = 3e−t d) y ′ + 2ty = 1

5. A Maxwell viscoelastic material is one for which the stress T (t) and
the strain-rate r(t) satisfy

dT
T +τ = κr,
dt
where τ and κ are positive constants. By solving this equation for T ,
and assuming T0 = T (0), show that
t
κ
Z
T = T0 e −t/τ
+ e(s−t)/τ r(s)ds.
τ 0

6. The Bernoulli equation is w′ = p(t)w + q(t)wn , which is nonlinear if

n 6= 0, 1. What is significant is that it can be solved by making the
substitution w = y 1/(1−n) , which results in a linear equation for y(t).
This was discovered by Leibniz, although it is not clear he was aware
of the solution (2.20) for a linear equation [Parker, 2013].
20 Chapter 2. First-Order Equations

a) If w′ = w − 5w3 , where w(0) = 1, what IVP does y satisfy?

b) Solve the IVP for y, and then transform back to determine the
function w.
c) One of Bernoulli’s brothers solved the Bernoulli equation by as-
suming that w(t) = u(t)v(t), where u satisfies u′ = pu, for
u(0) = 1. Use this method to solve w′ = w − 5w3 , where
w(0) = 1. This approach is the precursor to what is now known
as the method of variation of parameters.

2.3 Modeling
The principal objective of the examples to follow is to show how a dif-
ferential equation is the mathematical consequence of the assumptions
about a physical system.

2.3.1 Mixing
Typical mixing problems involve a continuously stirred tank, as illustrated
in Figure 2.2. As an example, suppose that water, containing salt, is
flowing into a well-stirred tank. At the same time, the mixture in the
tank is flowing out. The goal is to determine how much salt is in the tank
as a function of t.

Figure 2.2. Schematic of a continuous stirred tank.

The quantities of interest in this problem are:

Q(t): This is the amount of salt in the tank at time t. If the volume of
water in the tank is V , and c is the concentration of salt in the water,
then Q = cV .

Rin : This is the rate that salt is flowing into the tank. If the incoming
volumetric flow rate is Fin , and cin is the concentration of salt in the
incoming water, then Rin = cin Fin .

Rout : This is the rate that salt is flowing out of the tank. If the outgoing
volumetric flow rate is Fout , then Rout = cFout .
2.3. Modeling 21

If the initial amount of salt in the tank is Q0 , then the resulting IVP for
Q is:
dQ
= Rin − Rout ,
dt
Q(0) = Q0 .

Example 1
Suppose that salt water, containing 1/2 lbs of salt per gal, is poured into
a tank at 2 gal/min. Also, the water flows out of the tank at the same
rate. If the tank starts out with 100 gal of water, with 10 lbs of salt per
gal, find a formula for the total amount of salt in the tank.
Setup

inflow: Since Fin = 2, and cin = 1/2, then Rin = 1.

outflow: Since the mixture flows out at 2 gal/min, then the volume of
water in the tank stays at 100 gal. Also, since Fout = 2 and c =
1
Q/100, then Rout = 50 Q.
t = 0: Given that at the start there are 10 lbs of salt per gal, Q(0) = 1000.

The resulting IVP for Q is:

dQ 1
= 1 − Q, (2.31)
dt 50
Q(0) = 1000. (2.32)

Note that because of the way the variables have been defined, Q is mea-
sured in pounds and t is measured in minutes.
Solution

Using separation of variables, or the integrating factor solution (2.21),

one finds that Q(t) = 50 + 950e−t/50 .

Question: What is the eventual concentration of salt in the tank?

Answer using solution: Since limt→∞ Q(t) = 50, then the eventual con-
centration is 50/V = 21 lbs/gal.
Answer using physical reasoning: The concentration in the tank will even-
tually be the same as the concentration for the incoming flow, and
so the answer is 21 lbs/gal.
Answer using math reasoning: It is possible to determine the eventual
concentration directly from the differential equation, without know-
ing the solution. How this is done is explained in Section 2.4 (also,
see Exercise 5 in that section). 
22 Chapter 2. First-Order Equations

Example 2
Salt water, containing 3 lbs of salt per gal, flows into a 50 gal drum at 2
gal/sec. If the drum initially contains 10 gal of pure water, find a formula
for Q as a function of t.
Comments about this problem: There is no outflow, so the volume of
water will increase. However, it’s a 50 gal drum, so eventually it will fill
and start running over. When this occurs there is outflow, at a rate equal
to the incoming rate. To account for this, the problem needs to be split
into two phases, one where the volume is increasing, and the second when
it is a constant.
Solution

Phase 1: In this case, Rin = 6, Rout = 0, and Q(0) = 0. The resulting

IVP is
dQ
=6
dt
Q(0) = 0

The solution is Q(t) = 6t. Also, the volume of water in the tank is
V = 10 + 2t. So, this solution for Q holds for V ≤ 50, which means that
t ≤ 20.

Phase 2: As before, Rin = 6. For the outflow, the rate is 2 gal/sec and
the concentration in the outflow is Q/50. This means that Rout = Q/25.
Now, this phase starts at t = 20, and the amount of salt in the tank at
the start is 120 (this comes from the solution for Phase 1). This means
that the problem to solve is
dQ 1
= 6 − Q, for 20 < t,
dt 25
Q(20) = 120.

What is different about this problem is the time interval, which is not the
usual 0 ≤ t. However, this does not interfere with our solution methods,
and the solution can be found using an integrating factor or separation of
variables. One finds that the general solution of the differential equation
is
Q(t) = 150 + Ae−t/25 .
From the requirement that Q(20) = 120 it follows that A = −30e4/5 .

The Solution: Combining the Phase 1 and Phase 2 solutions, we have

that (
6t if 0 ≤ t ≤ 20,
Q(t) =
150 − 30e(20−t)/25 if 20 < t. 
2.3. Modeling 23

2.3.2 Newton’s Second Law

Suppose an object with mass m is moving along the x-axis. Letting
x(t) be its position, then its velocity is v = dxdt , and its acceleration is
d2 x dv
a = dt2 = dt . If the object is acted on by a force F , then from Newton’s
second law, which states that F = ma, we have that

dv
m = F. (2.33)
dt
What sort of differential equation this might be depends on how F de-
pends on v. Once (2.33) is solved for v, then the position is determined
by integrating the equation
dx
= v. (2.34)
dt
Typically, the initial velocity v(0) and initial position x(0) are given,
and these are used to determine the integration constants obtained when
solving the problem.

Vertical Motion
The object is assumed to be moving vertically, either up or down. In this
case, x(t) is the distance of the object from the ground. It is also assumed
that it is acted on by gravity, Fg , and a drag force, Fr . Consequently, the
total force is F = Fg + Fr . As for what these forces are:

Gravitational force: Assuming the gravitational field is uniform, then

Fg = −mg, where g is the gravitational acceleration constant. The minus
sign is because the force is in the downward direction.

Drag force: As long as the object is not moving very fast, the drag is
proportional to the velocity (see Exercise 10). In this case, Fr = −cv,
where c is a positive constant (the minus sign is because the force is in
the opposite direction to the direction of motion).

Units and Values: In the exercises, the value to use for g is usually stated.
If it is not given, then you should leave g unevaluated. Whatever value is
used, it is only approximate. If a more physically realistic value is needed,
then you should probably use the Somigliana equation. Finally, weight
is a force, so for an object that weighs w lbs, its mass can be determined
from the equation w = mg.

Example: Suppose a ball with a mass of 2 kg is dropped, from rest,

from a height of 1000 m. Assume that the forces acting on the object are
gravity, and a drag force due to air resistance, with c = 21 kg/s. Assume
that g = 10 m/s2 .
24 Chapter 2. First-Order Equations

Question 1: What is the resulting IVP for v, and what problem must be
solved to find x?
Answer: Since F = Fg + Fr = −mg − cv, where m = 2 and c = 1/2,
then from (2.33) the differential equation is
dv 1
= −10 − v. (2.35)
dt 4
Since the object is dropped from rest, then the initial condition is
v(0) = 0. Once v is known, then x is found by integrating (2.34),
and using the fact that x(0) = 1000. Also, note that v is measured
in meters per second, t is measured in seconds, and x in meters.

Question 2: What is the solution of the IVP, and the resulting solution
for x?
Answer: Using the integrating factor solution (2.21), it is found
that the general solution is v = −40 + ce−t/4 . Applying the initial
condition we get that

v = 40(−1 + e−t/4 ). (2.36)

Integrating x′ = 40(−1 + e−t/4 ), yields x = 40(−t − 4e−t/4 ) + c.

Since x(0) = 1000, then c = 1160. So, x = 40(−t − 4e−t/4 ) + 1160.

Question 3: What is the terminal velocity vT of the object?

Answer: The terminal velocity is defined as

vT = lim v(t).
t→∞

Consequently, from (2.36), we get that vT = −40 m/s. It is also

possible to determine vT without solving the IVP, and how this is
done is explained in Section 2.4.

Question 4: When does the object hit the ground?

Answer: It hits the ground when x = 0, which means that it is the
value of t that satisfies t + 4e−t/4 = 29. This can be solved using a
computer, but it is possible to obtain an approximate value fairly
easily. Assuming it takes several seconds to hit the ground, then
the 4e−t/4 term should be relatively small. For example, at t = 10,
4e−t/4 ≈ 0.3, and at t = 20, 4e−t/4 ≈ 0.03. Consequently, as an
approximation, we can replace the equation t + 4e−t/4 = 29 with
t = 29. In comparison, the numerically computed value is about
28.997 s. 

2.3.3 Logistic Growth or Decay

An assumption often made for the growth of the population of a species is
that the population grows at a rate proportional to the current population.
2.3. Modeling 25

If P (t) is the population at time t, then this assumption results in the

equation P ′ = kP . The solution is P (t) = P (0)ekt , which means that
there is exponential growth in the population. This is not sustainable in
the real world, and it is more realistic to assume that the rate of growth
slows down as the population increases. In fact, if the population is very
large, the population should decrease instead

of increase. A simple model
P
for this is to assume that k = r 1 − N , where r and N are positive
constants. The resulting differential equation is

′ P

P =r 1− P, (2.37)
N
which is known as the logistic equation. This nonlinear equation can be
solved using separation of variables, and partial fractions. Doing this, in
the case of when 0 < P < N ,
N dP
= rdt (2.38)
(N − P )P
⇒
1 1 
Z 
+ dP = rt + c
P N −P
⇒
P
ln = rt + c
N −P
⇒
P
= ert+c .
N −P
From this, we get
P = (N − P )c̄ert , (2.39)
where c̄ = ec is a positive constant. Doing the same thing for the case of
when N < P , one again gets (2.39) except that c̄ is a negative constant.
Moreover, for the divide by zero case of when P = 0, you get (2.39) but
c̄ = 0. In other words, except for when P = N , (2.39) holds with the
understanding that c̄ is an arbitrary constant. Solving (2.39) for P yields

N c̄ert
P = , (2.40)
1 + c̄ert
where c̄ is an arbitrary constant. If P (0) = P0 , and if P0 6= N , then
one finds that c̄ = P0 /(N − P0 ). When P (0) = N , this is a divide by
zero situation in (2.38), and the resulting solution is just the constant
P (t) = N .
The solution we have derived in (2.40) is known as the logistic function
or the logistic curve. When plotted for −∞ < t < ∞ it has a S, or
26 Chapter 2. First-Order Equations

P(0)
P-axis

0
0
t-axis

Figure 2.3. The logistic function (2.40), for −∞ < t < ∞, in the case of
when P (0) < N .

sigmoidal, shape as shown in Figure 2.3. It is one of those functions that

appears in so many applications that it deserves its own graph in this
textbook (hence Figure 2.3). 

2.3.4 Newton’s Law of Cooling

The assumption is that the rate of change of the temperature of an object
is proportional to the difference between its temperature and the ambient
temperature (i.e., the temperature of its surroundings). This is often
referred to as Newton’s law of cooling, but it also applies to heating an
object.
To write down the mathematical form of this statement, we introduce
the following:

T (t): This is the temperature of the object at time t.

Ta : This is the ambient temperature.
k: This is the proportionality coefficient.

If the initial temperature of the object is T0 , then the resulting IVP for
T is:
dT
= −k(T − Ta ), (2.41)
dt
T (0) = T0 . (2.42)

This problem can be solved using the integrating factor solution (2.21), or
by using separation of variables. It is found that T = Ta + (T0 − Ta )e−kt .

Example 1: Cooling a Cup of Coffee

According to the National Coffee Association, the ideal temperature for
brewing coffee is 200◦ F, and to get the most flavor out of it, you should
drink it when the coffee is between 120 and 140◦ F.
2.3. Modeling 27

Question 1: If the room temperature is 70◦ F, what is the solution of the

resulting IVP for T ?
Answer: Since T0 = 200 and Ta = 70, then

T = 70 + 130e−kt . (2.43)

Question 2: If the temperature is 180◦ F after 2 minutes, determine k.

Answer: From (2.43), 180 = 70 + 130e−2k . From this one finds that
1
k = 12 ln(13/11) min .

Question 3: When should you start drinking the coffee (according to the
National Coffee Association)?
Answer: The time when T = 140 occurs when 140 = 70 + 130e−kt ,
from which one finds that
ln(13/7)
t=2 min. (2.44)
ln(13/11)

Question 4: What is the computed value for the answer for Question 3?
Answer: It is t ≈ 7.4 minutes. 

Example 2: Nonlinear Cooling

Experimentally it has been observed that for certain fluids the k in (2.41)
is not constant. To account for this, according to what is known as the
Dulong-Petit law of cooling, the k in (2.41) is replaced with k(T − Ta )1/4 .
The resulting differential equation is
dT
= −k(T − Ta )5/4 .
dt
This requires cooling, and so it requires T ≥ Ta .

Question: As in Example 1, suppose that the room temperature is 70◦ F

and T (0) = 200◦ F. What is the solution of the resulting IVP?
Answer: Separating variables,
dT
− = kdt
(T − 70)5/4
⇒
4
= kt + c
(T − 70)1/4
⇒
4
(T − 70)1/4 = .
kt + c
28 Chapter 2. First-Order Equations

Solving this for T , we get that

 4 4
T = 70 + .
kt + c
Since T (0) = 200, then the above equation gives us that 130 =
(4/c)4 . Solving this we obtain c = 4/1301/4 . 

Reality Check : The models that are considered here are used to illustrate
how, and where, differential equations arise. As with all models, simplify-
ing assumptions are made to obtain the resulting mathematical problem.
Many of these assumptions are not considered or accounted for in our
examples, and the same is true for the exercises. As a case in point, New-
ton’s Law of Cooling is usually limited to cases of when |T − Ta | is not
very large, and its applicability depends on whether the heat flow is due
to conduction, convection, or radiation. Said another way, if you want to
impress your family at Thanksgiving by using the solution of the cook-
ing a turkey exercise (see below), just make sure to check on the turkey
temperature regularly to make sure your predictions are correct.

Exercises
In answering
√ the following questions, do not numerically evaluate numbers
2
such as 2, π/3, e , ln(4/3), etc. The exception to this is when the
question explicitly asks you to compute the answer.

1. The IVP for radioactive decay was derived in Example 1, on page 1.

a) What is the solution of the IVP for a radioactive material?
b) If 12 mg of a radioactive material decays to 9 mg in one day, find k.
c) The half-life of a radioactive material is the time required for it to
reach one-half of the original amount. What is the half-life of the
material in part (b)?
2. Radiocarbon dating uses the decay of carbon-14 to estimate how long
ago something died. The assumption is that the amount of carbon-
14 satisfies the radioactive decay problem derived in Example 1, on
page 1.
a) What is the solution of the IVP for a radioactive material?
b) The half-life of a radioactive material is the time required for it to
reach one-half of the original amount. The half-life of carbon-14 is
5,730 years. Use this to determine k.
c) The amount of carbon-14 is the same in all living organisms. When
an organism dies the amount starts to undergo radioactive decay.
Exercises 29

So, for radioactive dating you know N0 , as well as the current value
of N . Explain how knowing N0 , N , and k can be used to determine
t (which is the time that has passed since the organism died).
d) Measurements in 1991 determined that the amount of carbon-14
in the Temple Scroll, which is one of the Dead Sea scrolls found
at Qumran, to be 186.18. The amount in living organisms is 238.
Determine (i.e., compute) what two years the scroll could have been
written in. Note that in the BC/AD system there is no year zero,
so it goes from 1 BC to 1 AD.
Comments: In this problem, the amount of carbon-14 refers to the
amount relative to carbon-12. Also, the organism is the parchment
from the scroll, and the testing is described in Bonani et al. [1992].
Mixing
3. A tank contains 100 L of salt water with a concentration of 2 g/L. To
flush the salt out, pure water is poured in at 4 L/min, and the mixture
in the tank flows out at the same rate.
a) What is the resulting IVP for the total amount Q(t) of salt in the
tank?
b) Solve the IVP determined in part (a).
c) How long does it take until the amount of salt in the tank is 1% of
its original amount?
4. A tank contains 20 L of fresh water. Suppose water, containing 14 g/L
of salt, starts to flow into the tank at 2 L/min, and the well-stirred
mixture flows out at the same rate.
a) What is the resulting IVP for the amount Q(t) of salt in the tank?
b) Solve the IVP determined in part (a).
c) How much salt is in the tank after one hour?
5. Ten years ago, a factory started operation in a pristine valley. The
valley’s volume is 106 m3 . Each year the factory releases 105 m3 of
exhaust through its smoke stacks, and this exhaust contains 1000 kg of
pollutants. Assume that the well-mixed polluted air leaves the valley
at 105 m3 /yr.
a) What is the IVP for the amount of pollutant in the valley?
b) How much pollutant is in the valley now?
6. A small lake contains 60,000 gal of pure water. There is an inlet stream
of pure water into the lake, as well as an outlet stream, both flowing
at a rate of 100 gal/min. Suppose someone starts pouring water into
the lake at the rate of 10 gal/min that contains 5 lbs/gal of a chemical,
and they do this for 8 hours. While this happens the inlet stream of
pure water is unchanged, and the outflow rate from the lake remains
at 100 gal/min.
30 Chapter 2. First-Order Equations

a) What is the formula for the volume of the lake while the person is
pouring?
b) What IVP must be solved to determine the amount of the chemical
in the lake?
c) How much of the chemical is in the lake when the person stops
pouring?
d) Once the person stops pouring, what IVP must be solved to deter-
mine how much of the chemical is in the lake?

Newton’s Second Law

7. A mass of 10 kg is shot upward from the surface of the Earth with a
velocity of 100 m/s. In addition to gravity, assume that there is a drag
force Fr = −cv, where c = 5 kg/s. Assume that g = 10 m/s2 .
a) Write down the IVP for v, and then find its solution.
b) Find x.
c) How high does the object get?
8. A skydiver weighing 176 lbs drops from a plane that is at an altitude
of 5000 ft. Assume that g = 32 ft/s2 .
a) Before the parachute opens, the forces on the skydiver are gravity
and a drag force Fr = −cv. Assuming v(0) = 0, write down the
IVP for v, and then find the solution.
b) It is claimed that the terminal velocity of a person falling is −120 mph.
Use this to determine c.
c) If the parachute is opened after 10 s of free fall, what is the speed
of the skydiver when it opens?
d) Find the distance the skydiver falls before the parachute opens.
e) When the parachute is open, the drag force increases by a factor
of 8 from the free fall drag force. What is the resulting terminal
velocity of the skydiver?
9. A spherical object sinking to the bottom of a lake is acted on by three
forces: a drag force Fr = −cv, a buoyant force Fb , and gravity Fg .
According to Archimedes’ principle, the buoyant force is equal to the
weight of the water that is displaced by the sphere.
a) What is the formula for Fb in terms of the sphere’s radius a, the
water density ρ, and g?
b) The differential equation for the velocity of the sphere has the form
mv ′ = A − cv. What is A?
c) Assuming the sphere is released from rest, solve the resulting IVP
for v.
d) Find a formula for the terminal velocity in terms of c, a, ρ, and g.
What condition must be satisfied if the sphere is sinking?
Exercises 31

e) Assume the object is released a distance L from the bottom of the

lake. Also assume that it takes a while for it to hit the bottom. Use
an approximation similar to the one used in Question 4 on page 24
to derive an approximate formula for the time it takes it to hit the
bottom.

104

100
Drag Force

10-4

10-8 Experimental
F
r
10-12
10-6 10-4 10-2 100 102 104
Speed (m/sec)

Figure 2.4. Drag force on a smooth sphere as a function of the speed [Roos
and Willmarth, 1971, NASA, 2020]. The function Fr is used in Exercise 10.

10. A spherical object falling in the atmosphere is acted on by gravity, Fg ,

and a drag force Fr . It is assumed that Fr = −cv(1 − βv), where v is
the velocity. Both c and β are positive constants.
a) Assuming the sphere is dropped from rest, what is the resulting
IVP for v?
b) Solve the IVP for v.
c) Find a formula for the terminal velocity in terms of m, c, β, and g.
d) The constants in Fr are c = 6πRµ and β = Rρ/(9πµ), where R is
the radius of the sphere, ρ is the air density, and µ is the air viscosity.
For a baseball falling in the atmosphere, R = 0.037, µ = 1.8 × 10−5 ,
and ρ = 1.2 (using kg, m, s units). Also, m = 0.14 and assume that
g = 9.8. Compute the terminal velocity. How does this compare to
what is the speed of a typical fastball in professional baseball?
Comment: The drag force used in this problem is close to what is
observed experimentally. To demonstrate this, the experimentally
determined values of the drag, and the values determined using Fr ,
are shown in Figure 2.4 as a function of the speed |v|. This data
also shows that the assumption Fr = −cv is only valid if the speed
is no more than about 10−2 m/s.

Logistic Growth or Decay

11. It is often found that a population will grow exponentially if the popu-
lation is very small, and it will decrease exponentially if the population
is very large. A model for this is due to Beverton and Holt, and the
32 Chapter 2. First-Order Equations

equation to solve is
P
1− N
P′ =r P
P,
1+ N
where r and N are positive constants.
a) Assuming that P (0) = 12 N , solve the resulting IVP for P .
b) What is the limiting population P (∞) = limt→∞ P (t)?
12. The population of fish in a large lake can be modeled using the logistic
equation. If, in addition, the fish are caught at a constant rate h, the
equation for the population becomes
 P
P′ =r 1− P − h,
N
where r and N are positive constants. In this problem take r = 4,
h = 750, and N = 1000. Also, P (0) = 1000.
a) Solve the IVP for P .
b) What is the limiting population? In other words, what is P (∞) =
limt→∞ P (t)?

Cooling or Heating
13. Suppose coffee has a temperature of 200◦ F when freshly poured, and
the room temperature is 72◦ F. In this exercise use Newton’s law of
cooling.
a) What IVP does the temperature of the coffee satisfy?
b) What is the solution of the IVP?
c) If the coffee cools to 136◦ F in five minutes, what is k?
d) When does the coffee reach a temperature of 150◦ F?
14. Redo the previous exercise but use the Dulong-Petit law of cooling.
15. To cook a turkey you are to put it into a 350◦ F oven, and cook it
until it reaches 165◦ F. In answering the following questions, assume
Newton’s law of cooling is used.
a) Suppose the turkey starts out at room temperature, which is 70◦ F.
What IVP does the temperature satisfy?
b) Suppose that after two hours in the oven, the temperature of the
turkey is 140◦ F. How much longer before it is done?
c) Suppose the turkey is taken from the refrigerator, which is set to
40◦ F, and put directly into the oven. How much longer does it take
to cook than when the turkey starts out at room temperature? The
value for k is the same as in part (b).
16. A homicide victim was discovered at 1 p.m. in a room that is kept at
70◦ F. When discovered, the temperature of the body was 90◦ F, and
one hour later it had dropped to 85◦ F.
2.4. Steady States and Stability 33

a) Assuming Newton’s Law of Cooling, and normal body temperature

is 98.6◦ F, how long had the person been dead when the body was
discovered?
b) Compute the time of death. Round your answer so it just gives the
hour and minute (e.g., 7:13 a.m. or 5:32 p.m.).
17. Suppose that in Newton’s Law of Cooling that k is found to depend
on temperature. A common assumption is that k = k0 + k1 (T − Ta ),
where k0 and k1 are positive constants.
a) What is the resulting differential equation for T ?
b) To find T it makes things easier to introduce the variable S(t) =
T (t) − Ta . Rewrite the differential equation in part (a) in terms of
S. Also, if T (0) = T0 , what is S(0)?
c) Solve the resulting IVP in part (b) for S, and then use this to show
that
k0 ce−k0 t
T = Ta + ,
1 − k1 ce−k0 t
where c = S0 /(k0 + k1 S0 ) and S0 = T0 − Ta .
d) Using (2.43), it was found you have to wait about 7.4 minutes to
drink the coffee. Taking k0 = 21 ln(13/11) and k1 = 0.01, compute
how long you need to wait using the solution for T from part (c).

2.4 Steady States and Stability

All of the applications considered in the previous section have one thing in
common: the solution eventually approaches a constant value, or steady
state. This is not unusual, as this is what often happens. What is of
interest here is whether it is possible to determine the eventual steady
state without actually having to solve the problem.
To illustrate, as explained in the previous section, the population P (t)
of a species is determined by solving

P ′ = f (P ), (2.45)

where, for this example, we will take

f (P ) = 2(3 − P )P. (2.46)

The solution of this equation is given in (2.40), and it is plotted in Figure

2.5 for the case of when P (0) = 0.1, and when P (0) = 4.5. It shows that
for both initial values, the population approaches, asymptotically, P = 3.
In both cases the approach is monotonic, either increasing or decreasing.
What is important for this discussion is that it is possible to determine
the general behavior of the solution seen in Figure 2.5 without solving the
problem. This requires the following three observations:
34 Chapter 2. First-Order Equations

0
0 0.5 1 1.5 2

Figure 2.5. Solution of (2.45) and (2.46) in the case of when P (0) = 0.1,
and when P (0) = 4.5. The dashed red lines are the steady state values.

Steady States: If the solution does asymptotically approach a constant

value P , then P = P must be a solution of the differential equation.
This means that it is required that f (P ) = 0. From this and (2.46)
we get the two values P = 0 and P = 3. These are called steady
states for this equation.

Unstable: Even though the initial value P (0) = 0.1 is close to the steady
state P = 0, the solution moves away from P = 0. This happens
because of f (P ). To explain, the function f (P ) is plotted in Figure
2.6. It shows that f (P ) > 0 for 0 < P < 3. So, in this interval
P ′ (t) > 0, and this means that P is increasing. Similarly, since
f (P ) < 0 for 3 < P , then P is decreasing in this interval. The
arrows in Figure 2.6 indicate the corresponding movement of P . The
conclusion we derive from the arrows is that if P (0) is anywhere in
0 < P < 3, then the solution will move away from the steady state
P = 0. Because of this, the steady state is said to be unstable.

Stable: The second conclusion we make from the arrows in Figure 2.6 is
that if P (0) is anywhere in 0 < P < 3, then the solution increases

-3

-6
0 1 2 3 4 5

Figure 2.6. The function f (P ) in (2.46). The two steady states are shown
by the reds dots. The arrows indicate the direction P moves in the respective interval.
2.4. Steady States and Stability 35

towards the steady state P = 3. Moreover, if P (0) is anywhere in

3 < P , then the solution decreases towards the steady state P = 3.
A consequence of this is that, no matter what initial condition we
pick near P = 3,
lim P (t) = 3.
t→∞

For this reason, P = 3 is said to be an asymptotically stable

steady state.

The key to what guarantees that the steady state P = 3 is asymptot-

ically stable is that f (P ) is positive to the left of P = 3, and negative to
the right of it. In other words, f (P ) is a deceasing function at P = 3.
Consequently, if f ′ (3) < 0 then P = 3 is asymptotically stable. A similar
test can be made for an unstable steady state.

2.4.1 General Version

The reasoning used in the above example is easily extended to more gen-
eral differential equations. To do this, assume that the equation is

y ′ = f (y), (2.47)

where f ′ (y) is a continuous function of y. Because f (y) is assumed to

not depend explicitly on t, the equation is said to be autonomous. So,
y ′ = 1 + y 3 is autonomous, but y ′ = t + y 3 is not.

Steady State. y = Y is a steady state for (2.47) if it is constant and

f (Y ) = 0.

Stability Theorem. A steady state y = Y is asymptotically stable if

f ′ (Y ) < 0 and it is unstable if f ′ (Y ) > 0.

The idea underlying asymptotic stability is that if y(0) is any point close
to Y , then
lim y(t) = Y. (2.48)
t→∞

To explain this more mathematically, a steady state is either stable or

unstable. It is stable if you can control how far the solution gets from Y
by picking y(0) close to Y . Specifically, given any ε > 0, you can find a
δ > 0 so that |y(t) − Y | < ε if |y(0) − Y | < δ. If this is not possible then
Y is unstable. So, the steady state P = 0 in Figure 2.6 is unstable as it
is not possible to find P (0)’s near P = 0 that will result in the solution
staying near P = 0.
36 Chapter 2. First-Order Equations

In addition, a stable steady state is either asymptotically stable, which

means that the limit (2.48) holds, or it is said to be neutrally stable. The
latter occur, for example, for the steady states of y ′ = 0. Neutrally stable
steady states are not considered in this chapter but will be in Chapters 4
and 5.
The case of when y(0) = Y merits a comment. No matter if the steady
state is stable or unstable, if y(0) = Y , then y(t) = Y is a solution of the
resulting IVP. Consequently, what is of interest is what the solution does
if you start close, but not exactly at, a steady state.

Example 1: Find the steady states, and determine their stability, for

y ′ = y 2 − y − 6.

Answer: The steady states are found by solving y 2 − y − 6 = 0, and

from this we get Y = 3 and Y = −2. To determine their stability,
since f (y) = y 2 − y − 6, then f ′ (y) = 2y − 1. Since f ′ (3) = 5 > 0,
then Y = 3 is unstable, and since f ′ (−2) = −5 < 0, then Y = −2
is asymptotically stable. 

2.4.2 Sketching the Solution

As demonstrated in the above example, the stability theorem makes it is
relatively simple to determine if a steady state is stable or unstable. It
is also relatively easy to sketch the solution, and the following example
illustrates how this is done. Moreover, as you will see, this is done using
nothing more than the sketch of the function f (y).

Example 2: Sketch the solution of y ′ = f (y), where f (y) is given in

Figure 2.7.

Answer: To do this it is necessary to know y(0). Before picking this value,

we first see what can be determined about the solution.

Steady States: The steady states are the points where f (y) = 0. From
Figure 2.7, this happens when y = −2, y = 1, and y = 3. These are
identified using red dots in the figure. From the graph it is evident
that f ′ (−2) < 0 and f ′ (3) < 0, and this means that y = −2 and
y = 3 are asymptotically stable. Similarly, since f ′ (1) > 0, then
y = 1 is unstable.

Increasing or Decreasing: If f (y) > 0, then the solution is increasing, and

if f (y) < 0, then the solution is deceasing. The respective y intervals
where this happens are shown in Figure 2.7 using arrows.

We will now use the above conclusions to sketch the solution.

2.4. Steady States and Stability 37

-4 -3 -2 -1 0 1 2 3 4

Figure 2.7. The sketch of the function f (y) for Example 2.

y(0) = 1.3: This point is located between two steady states, specifically,
1 < y(0) < 3. According to Figure 2.7, y(t) increases monotonically
in this interval, and asymptotically approaches y = 3. A curve with
these properties is shown in Figure 2.8.
y(0) = 0.8: In this case, the point is located between two steady states,
namely, −2 < y(0) < 1. From Figure 2.7, y(t) decreases monoton-
ically in this interval, and asymptotically approaches y = −2. A
curve with these properties is shown in Figure 2.8.
y(0) = −4: For this initial condition, according to the information in Fig-
ure 2.7, y(t) increases monotonically, and asymptotically approaches
y = −2. A curve with these properties is shown in Figure 2.8. 

The sketching procedure outlined above leaves some things undeter-

-1

-2

-3

-4
0

Figure 2.8. Solution curves obtained using the information in Figure 2.7.
The dashed red lines are the steady state values.
38 Chapter 2. First-Order Equations

mined. For example, nothing was said about how steep the curves are,
or whether they are concave up or down. It is possible to determine this
using Figure 2.7, but this level of analysis is not considered in this text.

2.4.3 Parting Comments

A few closing comments about the material in this section are in order.

1. What is defined as a steady state here is sometimes called a critical

point, or an equilibrium point. Referring to them as a steady state
is consistent with what is used for time independent solutions of
partial differential equations.

2. The stability theorem does not cover the case of when f ′ (Y ) = 0.

However, the graphical method, as in Figure 2.6, can still be used.

3. When a solution moves away from an unstable steady state, it does

not necessarily approach the closest stable steady state. An example
of this is shown in Figure 2.8. Although Y = 3 is closer to the initial
point y(0) = 0.8, f (y) is negative for −2 < y < 1, and this means
the solution must decrease.

Exercises
1. For each equation, verify that Y = 0 is a steady state. Determine if it
is unstable or asymptotically stable.

a) y ′ = sin(1 − ey ) c) y ′ = −ey sin(y)

b) y ′ = y 5 − 3y 2 + y d) y ′ = (1 + y 9 ) ln(1 + y)

2. For each differential equation, find the steady states and determine if
they are asymptotically stable or unstable.

a) y ′ = y 2 + y − 2 c) y ′ = 4y − y 3 e) y ′ = y 4 − 3y 2 − 4
b) y ′ = y 3 − y d) y ′ = e−y − 2 f) y ′ = e2y − 4ey + 3

3. Sketch the solution curve for each of the given initial conditions.

a) y ′ = y 2 + y − 2 d) y ′ = e−y − 2
y(0) = −3; y(0) = 0 y(0) = 1; y(0) = 2
b) y ′ = y 3 − y e) y ′ = y 4 − 3y 2 − 4
y(0) = 3/4; y(0) = −1/4 y(0) = 1; y(0) = −3
c) y ′ = 4y − y 3 f) y ′ = e2y − 4ey + 3
y(0) = 1/2; y(0) = 3 y(0) = −1; y(0) = ln 2
Exercises 39

4. Sketch the solution of (2.47) based on the information provided. As-

sume that f (y) is zero only at y = −2 and y = 3.

a) f (2) = −4, b) f ′ (−2) = 1, c) f ′ (3) = 2,

y(0) = 1 y(0) = 0 y(0) = −1

5. For the mixing problem given in (2.31), (2.32), sketch the solution
without using the formula for the solution. Make sure to explain how
you do this.
6. For the population problem in Exercise 11, on page 31, sketch the
solution without using the formula for the solution. Use this to answer
part (b) of that exercise.
7. For the drag on a sphere, as described in Exercise 10 on page 31, de-
termine the terminal velocity without solving the IVP. In other words,
answer part (c) using only part (a) of that exercise.
√
8. This problem concerns solving y ′ = − 1 + y, where y(0) = 0.
a) Using separation of variables, what is the solution? It helps to
note, from the differential equation, that y ′ (0) = −1.
b) Using the method outlined in Example 2, sketch the solution.
c) Sketch the solution you found in part (a). Assuming your sketch
in part (b) is correct, is there anything wrong with your solution
in part (a)? If so, how should it be modified? Does your sketch
from part (b) need to be modified as well?
9. The population of fish in a lake can be modeled using the logistic
equation. However, assuming that the fish are caught at a constant
rate h, the equation for the population becomes
 P
P′ =r 1− P − h,
N
where r and N are positive constants.
a) Assuming that the loss due to fishing is small enough that 0 <
h < rN/4, find the two steady states for the equation. Label these
values as P1 and P2 , where P1 < P2 .
b) Determine whether P1 and P2 are unstable or asymptotically stable.
c) Letting f (P ) be the right hand side of the differential equation,
sketch f (P ) for 0 ≤ P < ∞. With this, answer the question in
Exercise 12(b) on page 32.
d) Assuming that P1 < P (0) < P2 , sketch the solution. Do the same
for the case of when P2 < P (0).
e) Sketch the solution if 0 < P (0) < P1 . In doing this remember that
P (t) can not be negative. Note that you will find that there is a
time te where extinction occurs, and the differential equation does
not apply to the fish population for te < t.
40 Chapter 2. First-Order Equations

10. The solution of a differential equation is shown in Figure 2.9. Explain

why it can not be the plot of the solution of the following differential
equations. You only need to provide one reason (even though there
might be several).

a) y ′ = 1 + y 2 c) y ′ = (y − 4)(y − 3)(y − 1)
b) y ′ = y − 4 d) y ′ = (y − 2)(4 − y)

0
0 1 2 3 4 5 6 7 8

Figure 2.9. Plot used in Exercise 10. The starting point is y(0) = 2.

11. The following refer to the solution of (2.47), where f (y) is continuous.
Sketch a function f (y) so the stated conditions hold. Make sure to pro-
vide a short explanation of why your function satisfies the conditions
stated. If it is not possible to find such a function, explain why.
a) The solution is strictly monotone increasing for y < 0, is strictly
monotone decreasing for y > 0, and there are no steady states.
b) The only asymptotically stable steady state is Y = 0, and the only
unstable steady states are Y = −1 and Y = 1.
c) The only asymptotically stable steady state is Y = 0, and the only
unstable steady states are Y = 1 and Y = 2.
12. This problem concerns what is known as one-sided stability, or semi-
stability. The differential equation considered is

y ′ = 2(3 − y)2 .

a) Show that there is one steady state Y , and f ′ (Y ) = 0.

b) Sketch f (y) for −∞ < y < ∞. Use this to explain why, except
when y = Y , y is an increasing function of t.
c) Using the same reasoning as for the population example, explain
why, if y(0) < Y , then limt→∞ y(t) = Y . However, if y(0) > Y ,
then limt→∞ y(t) = ∞.
d) Use the results from part (c) to explain why this is an example of
one-sided stability.
Chapter 3

Second-Order Linear
Equations

The general version of the differential equations considered in this

chapter can be written as

d2 y dy
2
+ p(t) + q(t)y = f (t), (3.1)
dt dt

where p(t), q(t), and f (t) are given. One of the reasons this equation gets
its own chapter is Newton’s second law, which, if you recall, is F = ma.
To explain, if y(t) is the displacement, then the acceleration is a = y ′′ ,
and this gives us the differential equation my ′′ = F . In this chapter we
are considering problems when F is a linear function of velocity y ′ and
displacement y. Later, in Chapter 5, we will consider equations where the
dependence is nonlinear. It is because of the connections with the second
law that f (t) in (3.1) is often referred to as the forcing function.
In the previous chapter, for first-order linear differential equations, we
very elegantly derived a formula for the general solution. This will not
happen for second-order equations. All of the methods derived in this
chapter are, in fact, just good, or educated, guesses on what the answer
is. There are non-guessing methods, and one example involves using a
Taylor series expansion of the solution. An illustration of how this is
done can be found in Exercise 8 on page 51.
To use a guessing approach, it becomes essential to know the math-
ematical requirements for what can be called a general solution. This is
where we begin.

Introduction to Differential Equations, M. H. Holmes, 2020

41
42 Chapter 3. Second-Order Linear Equations

3.1 Initial Value Problem

A typical initial value problem (IVP) consists of solving (3.1), for t > 0,
with the initial conditions

y(0) = y0 , and y ′ (0) = y0′ , (3.2)

where y0 and y0′ are given numbers. Given that our solution methods
involve guessing, it is important that we know when to stop guessing and
conclude we have found the solution. This is why the next result is useful.

Existence and Uniqueness Theorem. If p(t), q(t), and f (t) are con-
tinuous for t ≥ 0, then there is exactly one smooth function y(t) that
satisfies (3.1) and (3.2).

In stating that y(t) is a smooth function, it is meant that y ′′ (t) is defined

and continuous for t ≥ 0. Those interested in the proof of this theorem,
or the theoretical foundations of the subject, should consult Coddington
and Carlson [1997].
So, according to the above theorem, if we find a smooth function that
satisfies the differential equation and initial conditions, then that is the
solution, and the only solution, of the IVP.

3.2 General Solution of a Homogeneous Equation

The associated homogeneous equation for (3.1) is

d2 y dy
2
+ p(t) + q(t)y = 0. (3.3)
dt dt
We need to spend some time discussing what it means to be the general
solution of this equation. So, consider Exercise 5(a), in Section 1.2. As-
suming you did this exercise, you found that given solutions y1 = e2t and
y2 = et of y ′′ − 3y ′ + 2y = 0, then

y(t) = c1 y1 (t) + c2 y2 (t) (3.4)

is a solution for any value of c1 and c2 . What is important here is that this
is a general solution of the differential equation. As in the last chapter,
this means that any, and all, solutions of the differential equation are
included in this formula.
This gives rise to the question: what is required so a solution like the
one in (3.4) can be claimed to be a general solution? The key to answering
this is the uniqueness guaranteed by the above theorem. The specifics of
the analysis are not needed here. What is needed is the conclusion, which
is stated next.
3.2. General Solution of a Homogeneous Equation 43

General Solution. The function y = c1 y1 (t) + c2 y2 (t), where c1 and c2

are arbitrary constants, is a general solution of (3.3) if the following are
true:
1. y1 and y2 are solutions of (3.3), and
2. y1 and y2 are linearly independent.
Stating that y1 and y2 are linearly independent means that the only
constants c1 and c2 that satisfy
c1 y1 (t) + c2 y2 (t) = 0, ∀ t ≥ 0, (3.5)
are c1 = 0 and c2 = 0. This is, effectively, the same definition of linear
independence used in linear algebra. The difference is that we have func-
tions rather than vectors. If it is possible to find either c1 6= 0 or c2 6= 0 so
(3.5) holds, then y1 and y2 are said to be linearly dependent. Finally,
in (3.5), the symbol ∀ is a mathematical shorthand for “for all” or “for
every.”
Given two solutions y1 and y2 of (3.3), the easiest way to determine if
they are independent is to use what is called the Wronskian. To explain,
the Wronskian of y1 and y2 is defined as
!
y1 y 2
W (y1 , y2 ) ≡ det ′ . (3.6)
y1 y2′
For those unfamiliar with determinants, this can be written as
W (y1 , y2 ) ≡ y1 y2′ − y2 y1′ . (3.7)
The usefulness of this function is due, in part, to the next result.
Independence Test. If y1 and y2 are solutions of (3.3), then y1 and y2
are independent if, and only if, W (y1 , y2 ) is nonzero.

To explain how the Wronskian comes into this problem, (3.5) must hold
on the interval 0 ≤ t < ∞. So, (3.5) can be differentiated, which gives
us the equation c1 y1′ + c2 y2′ = 0. This, along with (3.5), provides two
equations for c1 and c2 . It is not hard to show that if W (y1 , y2 ) 6= 0, then
the only solution to these two equations is c1 = c2 = 0. Consequently, y1
and y2 are independent. The rest of the proof, along with some additional
information, can be found in Exercises 5 and 6.
Example: Show that y = c1 e−3t + c2 et is a general solution of y ′′ + 2y ′ −
3y = 0.
Answer: In this case, y1 (t) = e−3t and y2 (t) = et . It is not
hard to show that they are solutions of the differential equation
(see Section 1.2). To check on independence, from (3.7), W =
e−3t et − (−3)e−3t et = 4e−2t . This is not zero, and so the functions
are independent. Therefore, y is a general solution. 
44 Chapter 3. Second-Order Linear Equations

In the remainder of the chapter, except in Section 3.11 and in some of

the exercises for Section 3.9, we will only consider differential equations
of the form
d2 y dy
2
+ b + cy = f (t), (3.8)
dt dt
where b and c are given constants. The reasons are that these are easier
to solve, and, more importantly, they are also the most common second-
order differential equations that arise in applications.

Exercises
1. Assuming b 6= 0, show that y1 = 1 and y2 = e−bt are independent
solutions of y ′′ + by ′ = 0.
2. Assuming ω 6= 0, show that y1 = cos(ωt) and y2 = sin(ωt) are inde-
pendent solutions of y ′′ + ω 2 y = 0.
3. Assuming ω 6= 0, show that y = c1 eωt + c2 e−ωt is a general solution of
y ′′ − ω 2 y = 0.
4. Show y = c1 e−αt + c2 te−αt is a general solution of y ′′ + 2αy ′ + α2 y = 0.
d
5. If y1 and y2 are solutions of (3.3), show that dt W + p(t)W = 0. Use
this to derive Abel’s formula, which is that
Rt
p(r)dr
W (y1 , y2 ) = ce− 0 ,

where c is a constant.
6. Let y1 = (t − 1)2 and y2 = −(t − 1)|t − 1|.
a) On the same axes, sketch y1 and y2 for 0 ≤ t < ∞.
b) Use (3.5) to show that y1 and y2 are linearly independent for 0 ≤
t < ∞. Hint: Consider t = 0 and t = 2.
c) Show that W (y1 , y2 ) = 0. Explain why this, together with the result
in part (b), does not contradict the Independence Test.

3.3 Solving a Homogeneous Equation

The solution of the homogeneous equation

d2 y dy
+ b + cy = 0 (3.9)
dt2 dt

can be found by assuming that y = ert . With this, y ′ = rert , and y ′′ =

r2 ert , and so (3.9) becomes (r2 + br + c)ert = 0. Since ert is never zero,
we conclude that
r2 + br + c = 0. (3.10)
3.3. Solving a Homogeneous Equation 45

This is called the characteristic equation for (3.9). It is easily solved

using the quadratic formula, which gives us that

1 p

r= − b ± b2 − 4c . (3.11)
2

There are three possibilities here:

1. there are two real-valued r’s: this happens when b2 − 4c > 0,

2. there is one r: this happens when b2 − 4c = 0, and
3. there are two complex-valued r’s: this happens when b2 − 4c < 0.

The case of when the roots are complex-valued requires a short introduc-
tion to complex variables, and so it is done last.

3.3.1 Two Real Roots

When there are two real-valued roots, say, r1 and r2 , then the two cor-
responding solutions of (3.9) are y1 = er1 t and y2 = er2 t . It is left as an
exercise to show they are independent. Therefore, the resulting general
solution of (3.9) is
y = c 1 e r1 t + c 2 e r2 t .

3.3.2 One Real Root and Reduction of Order

When there is only one root, the second solution can be found using
what is called the reduction of order method. To explain, if you know a
solution y1 (t), it is possible to find a second solution by assuming that
y2 (t) = w(t)y1 (t). In our case, we know that y1 (t) = ert , where r =
−b/2, is a solution. So, to find a second solution it is assumed that
y(t) = w(t)ert . Substituting this into (3.9), and simplifying, yields the
differential equation

w′′ + (2r + b)w′ + r2 + br + c = 0.

Since r = −b/2, and 4c = b2 , then the above differential equation reduces

to just w′′ = 0. Integrating this once gives w′ = d1 and then integrating
again yields w = d1 t + d2 , where d1 and d2 are arbitrary constants. With
this our second solution is y = (d1 t+d2 )ert . A solution that is independent
of y1 = ert is obtained by taking d1 = 1 and d2 = 0, which means that
y2 = tert . Therefore, the resulting general solution of (3.9) is

y = c1 ert + c2 tert .
46 Chapter 3. Second-Order Linear Equations

3.4 Complex Roots

An example of a differential equation that generates complex-valued roots
is
y ′′ + 4y ′ + 13y = 0. (3.12)
Assuming y = ert , we obtain the characteristic equation r2 + 4r + 13 = 0.
The two solutions of this are r1 = −2 + 3i and r2 = −2 − 3i. Proceeding
as in the case of two real-valued roots, the conclusion is that the resulting
general solution of (3.12) is

y = c 1 e r1 t + c 2 e r2 t
= c1 e(−2+3i)t + c2 e(−2−3i)t . (3.13)

Because complex numbers are used in the exponents, if this expression is

used as the general solution, then c1 and c2 must be allowed to also be
complex-valued.
Although solutions as in (3.13) are used, particularly in physics, there
are other ways to write the solution that do not involve complex numbers.
Even if (3.13) is used, there is still the question of how to evaluate an
expression such as e3i . For this reason, a short introduction to complex
variables is needed.

3.4.1 Euler’s Formula and its Consequences

The key for working with complex exponents is the following formula.

Euler’s Formula. If θ is real-valued then

eiθ = cos θ + i sin θ. (3.14)

It is not possible to overemphasize the importance of this formula. It is

one of those fundamental mathematical facts that you must memorize.
For those who might wonder how this formula is obtained, it comes from
writing down the Maclaurin series of eiθ , cos θ, and sin θ, and then showing
that they satisfy (3.14).
As it must, (3.14) is consistent with the usual rules involving arith-
metic, algebra, and calculus. The examples below provide illustrations of
this fact.
√
Example 1: Since, by definition, i = −1, then i2 = −1, i3 = −i, and
i4 = 1. Also,

(a + ib)2 = (a + ib)(a + ib)

= a2 − b2 + 2iab.
3.4. Complex Roots 47

It is useful to be able to identify the real and imaginary part of a

complex number. So, if r = a + ib, and a and b are real, then

Re(r) = a, and Im(r) = b.

As an example, Re(5 − 16i) = 5, and Im(5 − 16i) = −16. Finally,

two complex numbers are equal only when their respective real and
imaginary
√ parts are equal. So, for example, to state that√ eiθ =
1 1
2 2(1 − i)√requires that, using Euler’s formula, cos θ = 2 2 and
sin θ = − 21 2. 

Example 2: eiπ = cos π + i sin π = −1.

This shows that the exponential function can be negative. More-
over, since eiπ = −1 then, presumably, ln(−1) = iπ (i.e., you can
take the logarithm of a negative number). This is true, but there are
complications related to the periodicity of the trigonometric func-
tions, and to learn more about this you should take a course in
complex variables. 

Example 3: eiπ/2 = cos π/2 + i sin π/2 = i. 

Example 4: Assuming θ and ϕ are real-valued, then

eiθ eiϕ = (cos θ + i sin θ)(cos ϕ + i sin ϕ)

= cos θ cos ϕ − sin θ sin ϕ + i(cos θ sin ϕ + sin θ cos ϕ)
= cos(θ + ϕ) + i sin(θ + ϕ)
= ei(θ+ϕ) . 

Example 5: Assuming r is real-valued, then

d irt d
e = (cos rt + i sin rt)
dt dt
= −r sin rt + ir cos rt
= ir(cos rt + i sin rt)
= ireirt . 

The next step is to extend Euler’s formula to a general complex num-

ber. With this in mind, let z = x + iy, where x and y are real-valued.
Using the usual law of exponents,

ez = ex+iy = ex eiy
= ex cos y + i sin y .


(3.15)

The above expression is what we need for solving differential equations.

48 Chapter 3. Second-Order Linear Equations

3.4.2 Second Representation

We return to the general solution given in (3.13). With (3.15), we get the
following

y = c1 e(−2+3i)t + c2 e(−2−3i)t
= c1 e−2t cos 3t + i sin 3t + c2 e−2t cos 3t − i sin 3t

= (c1 + c2 )e−2t cos 3t + i(c1 − c2 )e−2t sin 3t.

We have therefore shown that the general solution can be written as

y(t) = d1 e−2t cos 3t + d2 e−2t sin 3t. (3.16)

It is not difficult to check that the functions y 1 = e−2t cos 3t and y 2 =

e−2t sin 3t are solutions of (3.12), and they have a nonzero Wronskian.
Moreover, since y 1 and y 2 do not involve complex numbers, then d1 and
d2 in the above formula are arbitrary real-valued constants.

3.4.3 Third Representation

There is a third way to write the general solution that can be useful when
studying vibration, or oscillation, problems. This comes from making the
observation that given the values of d1 and d2 in (3.16), we can write them
as a point in the plane (d1 , d2 ). Using polar coordinates, it is possible to
find R and ϕ so that d1 = R cos ϕ and d2 = R sin ϕ. In this case,

y = d1 e−2t cos 3t + d2 e−2t sin 3t

= Re−2t cos ϕ cos 3t + sin ϕ sin 3t

= Re−2t cos(3t − ϕ). (3.17)

This last expression is the formula we are looking for. In this representa-
tion of the general solution, R and ϕ are arbitrary constants that satisfy
0 ≤ R, and 0 ≤ ϕ < 2π. The advantage of this form of the general
solution is that it is much easier to sketch the solution, and to determine
its basic properties. Its downside is that it can be a bit harder to find R
and ϕ from the initial conditions than the other two representations.

3.5 Summary for Solving a Homogeneous Equation

To solve
y ′′ + by ′ + cy = 0, (3.18)
where b and c are constants, assume y = ert . This leads to solving the
characteristic equation r2 + br + c = 0, and from this the resulting general
solution is given below.
3.5. Summary for Solving a Homogeneous Equation 49

Two Real Roots: r = r1 , r2 (with r1 6= r2 ).

y = c 1 e r1 t + c 2 e r2 t (3.19)

One Real Root: r = λ.

y = c1 eλt + c2 teλt (3.20)

Complex Roots: r = λ ± iµ (with µ 6= 0). Any of the following can be

used:

y = c1 e(λ+iµ)t + c2 e(λ−iµ)t , where c1 , c2 are complex-valued (3.21)

y = d1 eλt cos(µt) + d2 eλt sin(µt), where d1 , d2 are real-valued (3.22)

y = Reλt cos(µt − ϕ), where R ≥ 0, and 0 ≤ ϕ < 2π (3.23)

In what follows, (3.22) is used. The exception is in Section 3.10, where

(3.23) is preferred because it is easier to sketch.

Example 1: Find a general solution of y ′′ + 2y ′ − 3y = 0.

Answer: The assumption that y = ert leads to the characteristic
equation r2 + 2r − 3 = 0. The solutions of this are r = −3 and
r = 1. Therefore, a general solution is y = c1 e−3t + c2 et . 

Example 2: Find the solution of the IVP: y ′′ + 2y ′ = 0 where y(0) = 3

and y ′ (0) = −4.
Answer: The assumption that y = ert leads to the characteristic
equation r2 + 2r = 0. The solutions of this are r = −2 and r = 0.
Therefore, a general solution is y = c1 e−2t + c2 . To satisfy y(0) = 3
we need c1 + c2 = 3, and for y ′ (0) = −4 we need −2c1 = −4.
This gives us that c1 = 2, and c2 = 1. Therefore, the solution is
y = 2e−2t + 1. 

Example 3: Find the solution of the IVP: y ′′ − 2y ′ + 26y = 0 where

y(0) = 1 and y ′ (0) = −4.
Answer: The characteristic equation is r2 − 2r + 26 = 0, and the
solutions of this are r = 1 + 5i and r = 1 − 5i. Using (3.22), since
λ = 1 and µ = 5, the general solution has the form

y = d1 et cos(5t) + d2 et sin(5t).

To satisfy the initial conditions we need to find y ′ , which for our

solution is

y ′ = (d1 + 5d2 )et cos(5t) + (−5d1 + d2 )et sin(5t).

50 Chapter 3. Second-Order Linear Equations

So, to satisfy y(0) = 1 we need d1 = 1, and for y ′ (0) = −4 we need

d1 +5d2 = −4. This means that d2 = −1, and therefore the solution
of the IVP is y = et cos(5t) − et sin(5t). 

Example 4: Find the solution of the IVP: y ′′ − 9y = 0 where y(0) = −2

and y(t) is bounded for 0 ≤ t < ∞.
Answer: The assumption that y = ert leads to the quadratic equa-
tion r2 = 9. The solutions of this are r = −3 and r = 3. Therefore,
a general solution is y = c1 e−3t + c2 e3t . To satisfy y(0) = 1 we
need c1 + c2 = −2. As for boundedness, e−3t is a bounded function
0 ≤ t < ∞ but e3t is not. This means we must take c2 = 0. The
resulting solution is y = −2e−3t . 

As you might have noticed, in the above examples the formula for
the roots in (3.11) was not used. The reason is that it is much easier
to remember the way the characteristic equation is derived (by assuming
y = ert , etc) than by trying to remember the exact formula for the roots.

Exercises
π
1. Assuming that z1 = 1 + i, and z2 = e2+i 6 , find Re(z) and Im(z):

a) z = z1 − 8 c) z = z2 e) z = z1 z2
b) z = 2iz1 d) z = z1 + 4z2 f) z = (z2 )6

2. Assuming θ and ϕ are real-valued, show that the following hold:

1 f) ei(θ+2π) = eiθ
a) = −i
i
1 a − ib eiθ
b) = 2 g) ei(θ−ϕ) = iϕ
a + ib a + b2 e
eiθ 6= 0, ∀θ h) e dθ = −ieiθ + c
R iθ
c)
1
e−iθ = iθ 1
eiθ + e−iθ


d) i) cos θ = 2
e
1
(eiθ )2 = e2iθ eiθ − e−iθ


e) j) sin θ = 2i

3. Find the general solution of the given differential equation.

a) y ′′ + y ′ − 2y = 0 f) y ′′ − 6y ′ + 9y =0
b) 2y ′′ + 3y ′ − 2y = 0 g) 4y ′′ + 4y ′ + y =0
c) y ′′ + 3y ′ = 0 h) 4y ′′ + y = 0
d) 4y ′′ − y = 0 i) y ′′ − 2y ′ + 2y =0
e) y ′′ = 0 j) y ′′ + 2y ′ + 5y =0
Exercises 51

4. Find the solution of the IVP.

a) y ′′ − y ′ − 2y = 0, y(0) = 0, y ′ (0) = −1
b) 2y ′′ + 3y ′ − 2y = 0, y(0) = −1, y ′ (0) = 0
c) y ′′ + 3y ′ = 0, y(0) = −1, y ′ (0) = −1
d) 5y ′′ − y ′ = 0, y(0) = −1, y ′ (0) = −1
e) 3y ′′ − y = 0, y(0) = 3, y(t) is bounded for 0 ≤ t < ∞
f) y ′′ − 23 y ′ − y = 0, y(0) = 5, y(t) is bounded for 0 ≤ t < ∞
g) y ′′ + 2y ′ + y = 0, y(0) = −1, y ′ (0) = 0
h) y ′′ + 9y = 0, y(0) = −1, y ′ (0) = −1
i) y ′′ + 2y ′ + 5y = 0, y(0) = −1, y ′ (0) = −1
13
j) y ′′ − y ′ + 36 y = 0, y(0) = 2, y ′ (0) = 1

5. The roots of the characteristic equation are given. You are to find the
original differential equation (of the form given in (3.18)). If only one
value is given, that is the only root.

a) r = −1, 1 d) r = 0, 2 g) r = 2 ± 5i
b) r = 3, 5 e) r = 1 h) r = ±2i
c) r = ±2 f) r = 0

6. Use the method of reduction of order to find a second solution y2 (t),

and then show it is linearly independent of the given solution y1 (t).
a) (t + 1)2 y ′′ − 4(t + 1)y ′ + 6y = 0, y1 (t) = (t + 1)2
b) (t + 3)y ′′ − y ′ + 4(t + 3)3 y = 0, y1 (t) = sin(t2 + 6t)
c) (t + 1)y ′′ − (t + 2)y ′ + y = 0, y1 (t) = et
7. Answer the following questions by either providing one example show-
ing it is true, or explaining why it is not possible.
a) Is it possible to find values for b and c so that the solution of (3.18)
is such that limt→∞ y = 0, no matter what the initial conditions?
b) Is it possible to find values for b and c so that the solution of (3.18)
is a bounded function of t, no matter what the initial conditions?
c) Is it possible to find values for b and c so that the solution of (3.18)
is a periodic function of t, no matter what the initial conditions?
8. Suppose y(t) satisfies the IVP: y ′′ − 2y ′ + 2y = 0, where y(0) = −1 and
y ′ (0) = 0.
a) Without solving the IVP, determine y ′′ (0).
b) Without solving the IVP, determine y ′′′ (0), y ′′′′ (0), and y ′′′′′ (0).
c) Explain how it is possible to determine the Maclaurin series ex-
pansion of y(t) directly from the differential equation and initial
conditions.
52 Chapter 3. Second-Order Linear Equations

3.6 Solution of an Inhomogeneous Equation

We now turn to the problem of solving the inhomogeneous second-order
differential equation
d2 y dy
2
+ b + cy = f (t). (3.24)
dt dt
As with the homogeneous equation, the first task is to explain what form
a general solution will have.
Equation (3.24) shares a property with all linear inhomogeneous dif-
ferential equations. Namely, the general solution can be written as

y(t) = yp (t) + yh (t), (3.25)

where yp is a particular solution of the differential equation, and yh (t)

is the general solution of the associated homogeneous equation.
That the solution can be written in this way was discussed for linear
first-order equations in Section 2.2.1. As you recall, we had solved the
problem and then made the observation that the solution can be writ-
ten as in (3.25). For the second-order problems we are now considering,
the situation is reversed, and we will use (3.25) to construct the general
solution.
The associated homogeneous equation for (3.24) is just

d2 y dy
+ b + cy = 0. (3.26)
dt2 dt
How to find the general solution of this has been discussed in some detail,
and formulas for the solution are given in Section 3.5.
So, what remains is to determine how to find a particular solution of
(3.24). As you should recall, a particular solution is any function that
satisfies the differential equation. Since any function will do, we are not
really picky on how this function is determined. In fact, our go-to method
is nothing more than guessing what a particular solution might be. For
those who prefer a more systematic approach, an alternative method is
derived in Section 3.9. The guessing method, what is called the method
of undetermined coefficients, is considered first.

3.6.1 Non-Uniqueness of a Particular Solution

A particular solution is only required to be a solution of the differential
equation. It is possible, for any given differential equation, to have two
rather different looking functions both be particular solutions. As an
example, both y = 1 − t and y = 1 − t + 3et − 5e−2t are particular
solutions of y ′′ + y ′ − 2y = 4t. To explain what’s going on here, the
general solution of the differential equation is

y = 1 − t + c1 et + c2 e−2t ,
3.7. The Method of Undetermined Coefficients 53

where c1 and c2 are arbitrary constants. A particular solution of this

equation is a solution with particular choices for c1 and c2 . For the two
particular solutions given earlier, the first has c1 = c2 = 0 and the second
has c1 = 3 and c2 = −5.
For the most part, when trying to find a particular solution we will
be looking for the case of when c1 = c2 = 0.

3.7 The Method of Undetermined Coefficients

The objective is to be able to find a solution, any solution, that satisfies
d2 y dy
2
+ b + cy = f (t). (3.27)
dt dt
Depending on f (t), it is often possible to simply guess a solution. To
illustrate, suppose the equation to solve is

y ′′ + y ′ + 2y = 5e3t . (3.28)

This equation is asking for a function y, which if you differentiate it as

indicated, and add the results together you get 5e3t . A function that
will generate e3t in this way is e3t . In other words, it is reasonable to
expect that a particular solution will have the form y = Ae3t . Since
y ′ = 3Ae3t and y ′′ = 9Ae3t , then from the differential equation we require
that 14Ae3t = 5e3t . This will hold by taking A = 5/14, and therefore a
5 3t
particular solution is yp = 14 e .

Example 1: Find a particular solution of

y ′′ − 2y ′ + y = 2 cos 4t. (3.29)

Answer: The functions which will, if you differentiate them once or

twice, generate cos(4t) are cos(4t) and sin(4t). So, the assumption
is that a particular solution can be found of the form

y = A cos 4t + B sin 4t. (3.30)

Since y ′ = −4A sin 4t + 4B cos 4t, and y ′′ = −16A cos 4t − 16B sin 4t,
then (3.29) requires that

(−15A − 8B) cos 4t + (−15B + 8A) sin 4t = 2 cos 4t. (3.31)

Equating the coefficients of the cos 4t terms, and the coefficients of

the sin 4t terms, we get that −15A − 8B = 2 and −15B + 8A = 0.
Solving these two equations gives us that A = −30/289, and B =
−16/289. Therefore, a particular solution of (3.29) is
30 16
yp = − cos 4t − sin 4t.  (3.32)
289 289
54 Chapter 3. Second-Order Linear Equations

The key observation coming from the last example is that if you be-
lieve a function needs to be included in the guess for yp , then all of its
derivatives must be included. So, looking at (3.29) you would expect that
cos(4t) needs to be part of the guess, which means you must also in-
clude sin(4t). You do not need to include 4 sin(4t), or −4 sin(4t), because
sin(4t) is multiplied by an arbitrary constant in the guess (3.30), and this
can account for any constant factors that might be generated by taking
a derivative.
There are two situations when this guessing approach runs into trou-
ble. One is easily fixable and this is demonstrated in the next example.
The other situation is not fixable, and the cause of the difficulty is illus-
trated in Example 7 below.

Example 2: Find a particular solution of

y ′′ + 4y = 3 cos 2t.

Answer: Given what happened in the last example, you would ex-
pect that to find a particular solution you would assume that

y = A cos 2t + B sin 2t.

However, both cos 2t and sin 2t are solutions of the associated ho-
mogeneous equation. Because of this, the guess would give us that
y ′′ + 4y = 0, no matter what the values are for A and B. The fix is
to take the guess, and for the terms that are solutions of the associ-
ated homogeneous equation, multiply them by t. So, the modified
guess for this example would be

y = t(A cos 2t + B sin 2t).

To check that this works, since

y ′ = A cos 2t + B sin 2t + t(−2A sin 2t + 2B cos 2t),

and

y ′′ = 2(−2A sin 2t + 2B cos 2t) + t(−4A cos 2t − 4B sin 2t),

then from the differential equation we get

2(−2A sin 2t + 2B cos 2t) = 3 cos 2t.

Equating the coefficients of the cos 2t and sin 2t terms we get that
−4A = 0 and 4B = 3. Therefore, A = 0, B = 43 , and a particular
solution is yp = 43 t sin 2t. 
3.7. The Method of Undetermined Coefficients 55

When using the method of undetermined coefficients, the step that

requires the most thought is getting the guess correct. After that, it
is relatively straightforward to find the coefficients. Consequently, in the
examples below, only the appropriate guess is determined. In these exam-
ples, yh (t) is the general solution of the associated homogeneous equation,
and f (t) is the forcing function.

Example 3: What guess should be made for y ′′ − y ′ − 6y = t3 + 2?

Answer: Since f (t) = t3 + 2, then f ′ = 3t2 , f ′′ = 6t, and f ′′′ = 6.
So, a complete guess is y = At3 + Bt2 + Ct + D. It remains to
make sure that none of the functions in this guess is a solution of
the associated homogeneous equation. Since yh = c1 e3t + c2 e−2t ,
and the guess does not include e3t or e−2t , then our guess is, indeed,
complete. 

Example 4: What guess should be made for y ′′ − y ′ − 6y = te−5t ?

Answer: The initial guess is y = Ate−5t . However, y ′ = A(e−5t −
5te−5t ), and this includes a new function e−5t . This must be in-
cluded in the guess, and so a complete guess is y = Ate−5t + Be−5t .
Finally, since yh = c1 e3t + c2 e−2t , and the guess does not include e3t
or e−2t , then our guess is, indeed, complete. 

Example 5: What guess should be made for y ′′ − y ′ − 6y = 4t2 + 1 −

sin(πt)?
The guess for f (t) = 4t2 + 1 is y = A0 t2 + A1 t + A2 , and the guess
for f (t) = sin(πt) is y = B0 sin πt + B1 cos πt. So, for the equation
as given, a guess is

y = A0 t2 + A1 t + A2 + B0 sin πt + B1 cos πt.

Finally, since yh = c1 e3t + c2 e−2t , and the guess does not include e3t
or e−2t , then our guess is, indeed, complete. 

Example 6: What guess should be made for y ′′ + 4y ′ + 4y = 5e−2t ?

Answer: The initial guess is y = Ae−2t . However, for this equation,
yh = c1 e−2t + c2 te−2t and one of these functions appears in the
guess. The first modification y = Ate−2t also appears in yh , and
this means we need to multiply by t again. Therefore, the complete
guess is y = At2 e−2t . 

Example 7: What guess should you make if f (t) = ln(1 + t)?

Answer: The initial guess is y = A ln(1 + t). Its derivatives are
y ′ = A/(1 + y), y ′′ = −A/(1 + y)2 , y ′′′ = 2A/(1 + t)3 , etc. Unlike
the other examples, the list of different derivative functions does not
56 Chapter 3. Second-Order Linear Equations

if f (t) contains then yp (t) contains all of the following

eat eat

cos(ωt) or sin(ωt) cos(ωt), sin(ωt)

tn tn , tn−1 , · · · , 1

tn eat tn eat , tn−1 eat , · · · , eat

eat cos(ωt) or eat sin(ωt) eat cos(ωt), eat sin(ωt)

Table 3.1. Guesses when using the method of undetermined coefficients. Note
that the exponent n must be a non-negative integer. Also, adjustments are needed if
yp (t) contains a solution of the associated homogeneous equation (see Examples 2 and
6, and Example 3 in Section 3.8).

stop. In such cases, the method of undetermined coefficients should

not be used. So, the answer to the question is, there is no guess and
the method described in Section 3.9 should be used. 

3.7.1 Finding the Coefficients

In Example 1, we ended up with the equation

(−15A − 8B) cos 4t + (−15B + 8A) sin 4t = 2 cos 4t, ∀t ≥ 0. (3.33)

To find A and B we equated the coefficients of the cos 4t and sin 4t terms
in this equation. This can be done because these functions are linearly
independent, and this is explained below. This approach does not require
that you prove the functions are independent. Rather, if you think they
might be, and you then determine values for A and B so (3.33) is satisfied
based on this assumption, then you have found a particular solution.
The explanation of why linear independence can be used to determine
A and B starts with rewriting (3.33) as

c1 cos 4t + c2 sin 4t = 0, ∀t ≥ 0,

where c1 = −15A − 8B − 2 and c2 = −15B + 8A. According to the

definition of linear independence, as given in (3.5), if cos 4t and sin 4t are
independent then it must be that c1 = 0 and c2 = 0. In other words,
−15A − 8B = 2 and −15B + 8A = 0.

3.7.2 Odds and Ends

Most textbooks on differential equations have tables for various guesses
that you should make for the method of undetermined coefficients. The
3.8. Solving an Inhomogeneous Equation 57

fact is that they are mostly unreadable. It is much easier to just remember
the rules used in formulate the guess, and the earlier examples should be
reviewed for the particulars.
However, some do find a table useful, and one is provided in Table
3.1. A few comments need to be made about what is listed. First, if f (t)
contains tn , as well as tn−1 , or tn−2 , or tn−2 , etc, then the guess for tn
is all that you need (see Example 4 above). Second, when solving (3.9),
if one the functions in the left column is a solution of the associated ho-
mogeneous differential equation the guess must be modified. The needed
modification was explained earlier (see Examples 2 and 6).

3.8 Solving an Inhomogeneous Equation

As stated earlier, the general solution of
d2 y dy
2
+ b + cy = f (t), (3.34)
dt dt
can be written as
y(t) = yp (t) + yh (t), (3.35)
where yp is a particular solution, and yh is the general solution of the
associated homogeneous equation. We now know how to find yp and yh ,
and so we consider a few examples.

Example 1: Find a general solution of y ′′ − 3y ′ + 2y = 5t2 − 3.

Step 1: Find yh . The associated homogeneous equation is y ′′ −

3y ′ + 2y = 0. Assuming y = ert , one gets the characteristic equation
r2 − 3r + 2 = 0. The roots are r = 1 and r = 2, and so yh =
c1 et + c2 e2t .
Step 2: Find yp . The guess is y = At2 + Bt + C, which means
that y ′ = 2At + B and y ′′ = 2A. Inserting these into the differential
equation we get that
2At2 + (−6A + 2B)t + 2A − 3B + 2C = 5t2 − 3.
The coefficients of the respective tn terms on the left and right hand
sides must be equal. This means that:
t2 : 2A = 5
t1 : −6A + 2B = 0
t0 : 2A − 3B + 2C = −3 .
Solving, we get that A = 5/2, B = 15/2, and C = 29/4.
Step 3: The general solution is
5 15 29
y = t2 + t + + c1 et + c2 e2t . 
2 2 4
58 Chapter 3. Second-Order Linear Equations

Example 2: Find the solution of the IVP: y ′′ − 4y ′ + 5y = 10 − e3t where

y(0) = 3/2 and y ′ (0) = 0.

Step 1: Find yh . The associated homogeneous equation is y ′′ −

4y ′ + 5y = 0. Assuming y = ert , one gets the characteristic equation
r2 − 4r + 5 = 0. The roots are r = 2 ± i, and so yh = c1 e2t cos t +
c2 e2t sin t.
Step 2: Find yp . The guess is y = A + Be3t , which means that
y ′ = 3Be3t and y ′′ = 9Be3t . Inserting these into the differential
equation we get that
2Be3t + 5A = 10 − e3t .
Equating the coefficients of the respective functions, 2B = −1 and
5A = 10. Solving, we get that A = 2 and B = −1/2.
Step 3: The general solution is y = 2− 21 e3t +c1 e2t cos t+c2 e2t sin t.
Step 4: To satisfy y(0) = 3/2 we need 3/2+c1 = 3/2, so c1 = 0. For
y ′ (0) = 0 we need −3/2 + c2 = 0, giving c2 = 3/2. The conclusion
is that the solution of the IVP is
1 3
y = 2 − e3t + e2t sin t. 
2 2

Example 3: Find a general solution of y ′′ − 2y ′ = −3t2 .

Step 1: Find yh . The associated homogeneous equation is y ′′ −

2y ′ = 0. Assuming y = ert , one gets the characteristic equation
r2 − 2r = 0. The roots are r = 0 and r = 2, and so yh = c1 + c2 e2t .
Step 2: Find yp . The initial guess is y = At2 +Bt+C. However, one
of the terms in this guess is a solution of the homogeneous equation,
and so the guess must be modified to y = t(At2 + Bt + C). Inserting
this into the differential equation we get that
6At + 2B − 2(3At2 + 2Bt + C) = −3t2 .
Equating the coefficients of the respective powers of t, we get that
−6A = −3, 6A − 4B = 0, and 2B − 2C = 0. Solving yields A = 1/2,
B = 3/4, and C = 3/4.
Step 3: The general solution is therefore
1
y = t(2t2 + 3t + 3) + c1 + c2 e2t . 
4

Exercises
1. Find the general solution of the given differential equation.
Exercises 59

a) y ′′ − y ′ − 6y = 6et i) y ′′ − 2y ′ + 5y = 5t2 + 4
b) y ′′ + 3y ′ + 2y = sin πt j) y ′′ + 2y ′ + 10y = 3et + 1
c) y ′′ + 4y ′ − 5y = 2t2 k) y ′′ − 3y ′ = t3 − 6
d) 5y ′′ − y ′ = e−t + 3 cos 2t l) 3y ′′ + y ′ − 2y = 3e−2t − et
e) 3y ′′ − 5y ′ − 2y = t3 − 2t m)y ′′ − 8y ′ + 17y = e4t sin t
f) 8y ′′ − 2y ′ − y = 4 + 5 sin 2t n) y ′′ − 5y ′ − 6y = −3 sin(t + 7)
g) y ′′ + 4y = tet o) y ′′ + 3y ′ + 2y = sin2 t
h) y ′′ − 5y ′ − 6y = 10t sin(3t) p) 4y ′′ + y ′ = sin(t) cos(t)

2. Find the solution of the given IVP.

a) y ′′ + y ′ − 2y = 3t, y(0) = 0, y ′ (0) = 0
b) y ′′ + 4y = t2 , y(0) = 1, y ′ (0) = 0
c) y ′′ − y ′ = sin t, y(0) = 1, y ′ (0) = −1
d) y ′′ + 3y ′ = 2t, y(0) = 1, y ′ (0) = 0
e) y ′′ + 4y ′ + 4y = −3e2t , y(0) = 1, y ′ (0) = 0
f) 4y ′′ − y = e−t/2 + 1, y(0) = 0, y ′ (0) = 0
g) y ′′ + 9y = −2 sin(3t), y(0) = 0, y ′ (0) = 0
h) y ′′ + 2y ′ + 5y = e−t , y(0) = −1, y ′ (0) = 0
i) y ′′ − y ′ + 21 y = 25 cos 3t, y(0) = −1, y ′ (0) = 0
3. For the following, determine a complete guess that can be used to find
a particular solution (you do not need to find the coefficients).

a) y ′′ + y ′ − 2y = t5 − t2 g) y ′′ − y ′ + 6y = e−t cos 3t
b) y ′′ + 4y = t cos t h) 4y ′′ − y = −2(t − 1)7
c) y ′′ + 4y = t + sin 2t i) y ′′ − 2y ′ + 2y = et−5 cos t
d) y ′′ − y ′ = 1 + sin t j) y ′′ + 4y = cos(2t + 3)
e) y ′′ + 3y ′ = 1 + e−3t k) y ′′ + 25y = −3 sin(5t + 7)
R1√
f) y ′′ + y ′ + 2y = t3 e−2t l) y ′′ + y ′ + y = 0 s cos(2t − s)ds

4. The idea underlying undetermined coefficients has nothing to do with

the differential equation being second-order. What is required is a
linear differential equation with constant coefficients. As an example,
for the first-order equation y ′ +y = e2t you assume a particular solution
of the form y = Ae2t . The associated homogeneous equation is y ′ +y =
0, which means that yh = c1 e−t . Finding the solution in this way is
easier than using an integrating factor (which is the way it is done in
Example 1 of Section 2.2). Find the general solution of the following
first-order equations using the method of undetermined coefficients.
60 Chapter 3. Second-Order Linear Equations

a) y ′ − 6y = 2et e) y ′ − 4y = t sin 2t
b) 3y ′ + 2y = sin πt f) y ′ − 6y = te−t + 2
c) y ′ + 3y = 2t g) 3y ′ + 2y = e−t cos(t)
d) 5y ′ − y = e−t + 3t h) y ′ − 2y = cos(2t + 5)

3.9 Variation of Parameters

When the method of undetermined coefficients works, it is relatively easy
to use it to find a particular solution. However, as illustrated in Example
7 of Section 3.7, it does not always work. In such cases, the method of
variation of parameters can be used. Interestingly, this method is also
based on a guess. Namely, to find a particular solution of

d2 y dy
2
+ b + cy = f (t), (3.36)
dt dt
it is assumed that
y = u1 (t)y1 (t) + u2 (t)y2 (t), (3.37)
where y1 and y2 are independent solutions of the associated homogeneous
equation. As you should notice, the guess (3.37) resembles the general
solution of the associated homogeneous equation. However, instead of
arbitrary constants, there are now unknown functions u1 and u2 . Our job
is to find these functions. Although it might not appear to be significant
right now, we are looking for a single function, yp , yet our guess contains
two unknown functions. This means that we have the option to pick
one of these two functions anyway we wish. We will use this option to
advantage to find yp .
Our task is simple in that (3.37) must be a solution of (3.36). So, in
preparation for substituting (3.37) into (3.36) note that

y ′ = u′1 y1 + u1 y1′ + u′2 y2 + u2 y2′ .

We now use the option of picking u1 or u2 anyway we want. The specific

choice is that
u′1 y1 + u′2 y2 = 0. (3.38)
So y ′ = u1 y1′ + u2 y2′ , and y ′′ = u1 y1′′ + u′1 y1′ + u2 y2′′ + u′2 y2′ . Substituting
these into (3.36), we get that

u′1 y1′ + u′2 y2′ + u1 (y1′′ + by1′ + cy1 ) + u2 (y2′′ + by2′ + cy2 ) = f. (3.39)

Using the fact that y1 and y2 are solutions of the associated homogeneous
equation, then the above equation reduces to

u′1 y1′ + u′2 y2′ = f. (3.40)

3.9. Variation of Parameters 61

Therefore, to find u1 and u2 , we must solve (3.38) and (3.40). This is

fairly easy. First, from (3.38), u′2 = −u′1 y1 /y2 . Inserting this into (3.40)
we get that
(y2 y1′ − y1 y2′ )u′1 = y2 f.
This can be written as

−W (y1 , y2 )u′1 = y2 f,

where W (y1 , y2 ) is the Wronskian as defined in (3.6). Solving this gives

us that Z t
y2 (s)f (s)
u1 (t) = − ds. (3.41)
0 W (y 1 (s), y2 (s))
Inserting into (3.38), and integrating, we obtain
Z t
y1 (s)f (s)
u2 (t) = ds. (3.42)
0 W (y1 (s), y2 (s))

Therefore, the particular solution we have found is

Z t Z t
y2 (s)f (s) y1 (s)f (s)
yp (t) = −y1 (t) ds + y2 (t) ds.
0 W (y1 (s), y2 (s)) 0 W (y1 (s), y2 (s))
(3.43)

Example 1: Find a particular solution of y ′′ − 3y ′ + 2y = t using varia-

tion of parameters.

Step 1: Find y1 and y2 . The associated homogeneous equation is

y ′′ − 3y ′ + 2y = 0. Assuming y = ert , one gets the characteristic
equation r2 − 3r + 2 = 0. The roots are r = 1 and r = 2, and so
y1 = et and y2 = e2t .
Step 2: Find u1 . Since W = y1 y2′ − y2 y1′ = e3t , and f = t, then
from (3.41),
Z t 2s Z t
e s
u1 (t) = − 3s
ds = − se−s ds.
0 e 0

Using integration by parts yields u1 = (1 + t)e−t − 1.

Step 3: Find u2 . From (3.42), and using integration by parts,
Z t s Z t
e s 1
se−2s ds = 1 − (2t + 1)e−2t .

u2 (t) = 3s
ds =
0 e 0 4
Step 4: Collecting our results,
1
yp = (1 + t)e−t − 1 et + 1 − (2t + 1)e−2t e2t
  
4
1 3 1
= t + − et + e2t . 
2 4 4
62 Chapter 3. Second-Order Linear Equations

3.9.1 The Solution of an IVP

It is not hard to show that yp (t), given in (3.43), satisfies yp (0) = 0 and
yp′ (0) = 0. Therefore, the solution of the IVP

d2 y dy
2
+ b + cy = f (t),
dt dt

where y(0) = y0 and y ′ (0) = y0′ , is

y(t) = c1 y1 (t) + c2 y2 (t) + yp (t), (3.44)

where c1 and c2 are found by solving

c1 y1 (0) + c2 y2 (0) = y0 , (3.45)

c1 y1′ (0) + c2 y2′ (0) = y0′ . (3.46)

It needs to be remembered that yp (t) appearing in (3.44) is the particular

solution given in (3.43).
√
Example 2: Find the solution of the IVP: y ′′ + 4y = t, where y(0) = 1,
and y ′ (0) = 0.

Step 1: Find y1 and y2 . The associated homogeneous equation is

y ′′ + 4y = 0. Assuming y = ert , one gets the characteristic equation
r2 = −4. The roots are r = ±2i, and so y1 = cos 2t and y2 = sin 2t.
√
Step 2: Find u1 . Since W = y1 y2′ − y2 y1′ = 2, and f = t, then
from (3.41),
Z t
1√
u1 (t) = − s sin(2s)ds.
0 2

The answer is left as a definite integral because it is not possible to

express it in terms of elementary functions.
Step 3: Find u2 . From (3.42),
t
1√
Z
u2 (t) = s cos(2s)ds.
0 2

Step 4: Solve (3.45) and (3.46). One finds that c1 = 1 and c2 = 0.

Step 5: Therefore, from (3.43) and (3.44), the solution is

1 t√ 
Z

y = cos(2t) + s sin(2t) cos(2s) − cos(2t) sin(2s) ds
2 0
1 t√
Z


= cos(2t) + s sin 2(t − s) ds. 
2 0
Exercises 63

A couple of comments need to be made about (3.44). First, it is the

solution of the IVP, irrespective of what continuous function f (t) is used.
It can also be adapted so it is the solution for more general problems
(see Exercise 3). A drawback is that it can require more work to find
the solution. As a case in point, it is much easier to do Example 1
using undetermined coefficients. However, for Example 2, undetermined
coefficients does not work, and this means that (3.44) is the method of
choice. The recommendation is to first consider whether undetermined
coefficients can be used, as it is usually fairly simple to carry out. It
also has the advantage that it is easier to remember than the variation of
parameters formula.

Exercises
1. Using variation of parameters, find a particular solution of the given
differential equation.

a) 2y ′′ + 3y ′ − 2y = 25e−2t d) y ′′ + 3y ′ = t3/2 + 1
2
b) y ′′ − 2y ′ + 2y = 6 e) 5y ′′ − y ′ = 1+t
c) y ′′ + y ′ − 2y = 3 ln(1 + t) f) 4y ′′ − y = sin(1 + t2 )

2. Find the solution of the IVP where the differential equation comes
from the previous problem, and the initial conditions are y(0) = 1 and
y ′ (0) = 0.
3. The formula for a particular solution given in (3.43) applies to the more
general problem of solving y ′′ +p(t)y ′ +q(t)y = f (t). In this case, y1 and
y2 are independent solutions of the associated homogeneous equation
y ′′ + p(t)y ′ + q(t)y = 0. In the following, show that y1 and y2 satisfy
the associated homogeneous equation, and then determine a particular
solution of the inhomogeneous equation.
a) t2 y ′′ − t(t + 2)y ′ + (t + 2)y = 2t3 ; y1 (t) = t, y2 (t) = tet
b) ty ′′ − (t + 1)y ′ + y = t2 e2t ; y1 (t) = 1 + t, y2 (t) = et
c) t2 y ′′ − 3ty ′ + 4y = t5/2 ; y1 (t) = t2 , y2 (t) = t2 ln t
4. The Bessel equation of order p is t2 y ′′ + ty ′ + (t2 − p2 )y = 0. In this
problem assume that p = 21 .
√ √
a) Show that y1 = sin t/ t and y2 = cos t/ t are linearly independent
solutions for 0 < t < ∞.
b) Use the result from part (a), and the preamble in Exercise 3, to find
a particular solution of t2 y ′′ + ty ′ + (t2 − 1/4)y = t3/2 cos t.
64 Chapter 3. Second-Order Linear Equations

3.10 Linear Oscillator

The problem considered involves a mass, spring, and dashpot as illus-
trated in Figure 3.1. The differential equation in this case has the form

mu′′ + cu′ + ku = f (t), (3.47)

where f (t) is an external forcing function. In this equation, u(t) is the dis-
placement of the mass from its rest position, with positive in the upward
direction. The physical interpretation of the terms in this differential
equation, and the basic properties of the solution are described in the
following pages.

Figure 3.1. Mass-spring-dashpot system.

3.10.1 The Spring Constant

To begin, a spring of length ℓ is suspended as illustrated in Figure 3.2.
After this, an object with mass m is attached, which stretches the spring
a distance L. The forces on the object in this case are gravity, Fg , and
the restoring force from the spring, Fs . The gravitational force is just
Fg = −mg, where g is the gravitational acceleration constant. The spring
force is determined using Hooke’s law, which states that the restoring
force is proportional to how much the spring is stretched. To translate
this into a mathematical formula, according to Hooke’s law, Fs = kL,
where k is the proportionally constant, and it is referred to as the spring
constant. The whole system is at rest, and so, from Newton’s second
law, we have that Fs + Fg = 0. From this we obtain
mg
k= . (3.48)
L

3.10.2 Simple Harmonic Motion

Now, with the mass attached, we set it in motion. For example, as il-
lustrated in Figure 3.2, the mass is pushed up and then released. The
equation governing the motion is, again, determined from Newton’s sec-
ond law. As before, the gravitational force is Fg = −mg. From Hooke’s
law, the restoring force due to the spring is Fs = k(L − u), where u(t) is
the displacement of the mass from its rest position. Since F = ma, and
the force in this problem is F = Fg + Fs , we get the following differential
3.10. Linear Oscillator 65

Figure 3.2. Left: The original spring. Middle: The situation after the mass
is attached, and at rest. Right: Displacement u(t) of the mass from its rest position.

equation
mu′′ + ku = 0. (3.49)
To find the general solution of (3.49), from the assumption that u = ert
the characteristic equation is found to be mr2 + k = 0. This produces the
roots r = ±ω0 i, where r
k
ω0 = . (3.50)
m
From (3.23), the general solution can be written as

u = R cos(ω0 t − ϕ), (3.51)

where R ≥ 0, and 0 ≤ ϕ < 2π. This periodic function corresponds to

what is called simple harmonic motion. In this context, the coefficient
R is the amplitude, ω0 is the natural frequency, and the period is
T = 2π/ω0 . The argument θ = ω0 t − ϕ of the cosine is called the phase,
and −ϕ is the phase shift, or the phase constant.
In terms of initial conditions, it is usual to specify the initial displace-
ment and the initial velocity. Together, these correspond to

u(0) = u0 , and u′ (0) = u′0 , (3.52)

where u0 and u′0 are given. To satisfy these, from (3.51), we get that

R cos(ϕ) = u0 , (3.53)
u′0
R sin(ϕ) = . (3.54)
ω0

Figure 3.3. The initial conditions as expressed in (3.53) and (3.54), located
using the black dot, and the value of R and ϕ.
66 Chapter 3. Second-Order Linear Equations

Finding R: Using the identity cos2 (ϕ) + sin2 (ϕ) = 1, it follows that
s
 u ′ 2
0
R = u20 + . (3.55)
ω0

Finding ϕ: The value for ϕ depends on whether u0 and u′0 are positive or
negative, as illustrated in Figure 3.3. To compute ϕ, assuming that
u0 6= 0, the ratio of (3.54) with (3.53) yields
u′0
tan(ϕ) = .
ω0 u0
The principal value of the arctan function is denoted as Arctan, and it
satisfies − π2 < Arctan(z) < π2 . This is the value most calculators, or
programs like MATLAB, give when evaluating arctan(z). So, setting
z = u′0 /(ω0 u0 ),

 Arctan(z)
 if u0 > 0 and u′0 ≥ 0,
ϕ= Arctan(z) + π if u0 < 0, (3.56)
if u0 > 0 and u′0 < 0.

 Arctan(z) + 2π

If u0 = 0, then ϕ = π/2 if u′0 > 0, and ϕ = 3π/2 if u′0 < 0.

Example 1: Suppose the mass is set in motion by pulling it down 2 cm

and then releasing it with an upward velocity of 1 cm/s. Also, assume
that k = 1 kg/s2 and m = 9 kg.
Question 1: Find the resulting simple harmonic motion, and then sketch
the solution.
Answer: The initial conditions are u(0) = −2, and u′ (0) = 1. Also,
from (3.50), ω0 = 1/3, and this means, using (3.51), that the general
solution is u = R cos(t/3 − ϕ). From (3.55), the amplitude is R =
√
13. As for ϕ, from (3.56), ϕ = Arctan(−3/2) + π. The resulting
solution is shown in Figure 3.4. Note that the period T = 2π/ω0 =
6π.

Figure 3.4. Simple harmonic motion solution for Example 1.

3.10. Linear Oscillator 67

Figure 3.5. Phase function θ for Example 1.

Question 2: When is the first time that u(t) = R?

Answer: The solution can be written as u = R cos θ, where the
phase is θ = t/3 − ϕ. This linear function of t is shown in Figure
3.5, and it includes the fact that π/2 < ϕ < π. Now, cos θ = 1 when
θ = 0, ±2π, ±4π, · · · . According to Figure 3.5, the first possible
value is θ = 0 and this occurs when t = 3ϕ. This is labeled as t1 in
Figure 3.5. 

3.10.3 Damping
We will now include a damping mechanism. It is assumed that the damp-
ing force is proportional to the velocity. For the mass-spring system the
resistance is usually illustrated as a dashpot, as shown in Figure 3.1. Ir-
respective of exactly what mechanism is involved, the result is that the
damping force is Fd = −cv, where v = u′ is the velocity, and c is the
damping constant and it is non-negative. From the equation F = ma,
and the fact that F = Fs + Fg + Fd , the resulting differential equation is

mu′′ + cu′ + ku = 0. (3.57)

Finding the general solution is straightforward. Assuming u = ert ,

then the resulting characteristic equation is mr2 + cr + k = 0. The roots
are
1  p 
r= − c ± c2 − 4mk . (3.58)
2m
Just as in Section 3.5, there are three cases to consider. The only differ-
ence now is that certain terminology is introduced to identity the cases.
Over-damped: This means that the damping constant is large enough
that c2 > 4mk. In this case both roots are real-valued, and the resulting
general solution is
u = c 1 e r1 t + c 2 e r 2 t , (3.59)
1
√
1
√

where r1 = 2m − c + c2 − 4mk and r2 = 2m − c − c2 − 4mk . It
is worth noting that not only are the roots real, they are both negative.
Therefore, no matter what the initial conditions,

lim u = 0. (3.60)
t→∞
68 Chapter 3. Second-Order Linear Equations

Critically damped: This means that the damping constant has just
that right value that c2 = 4mk. So, there is one root, and the resulting
general solution is
u = (c1 + c2 t)ert , (3.61)
where r = −c/(2m). So, as for the previous case, no matter what the
initial conditions, (3.60) holds.

Under-damped: This means that the damping constant is small enough

that c2 < 4mk. The roots are complex-valued, and the resulting general
solution is
u = Reλt cos(µt − ϕ), (3.62)
√
where λ = −c/(2m), and µ = 4mk − c2 /(2m). The solution is not
periodic, but it is oscillatory with an amplitude Reλt that decays to
zero (assuming, of course, that c > 0). Consequently, no matter what
the initial conditions, (3.60) holds.

One conclusion coming from the above discussion is that because of

damping, no matter what the initial conditions, the solution decays ex-
ponentially to zero. The role of damping, and how it affects the solution,
is explored in the next examples.

Example 2: For a mass-spring-dashpot system, suppose that m = 2,

k = 1, and c = 1. Also, assume that the initial conditions are u(0) = 1,
and u′ (0) = 2.
Question 1: What is the solution? √
Answer: From (3.58), the roots are r = (−1 ± i √7)/4. So, this is a
case of under-damping, with λ = −1/4 and µ = 7/4. From (3.62),
the general solution is
1√
 
−t/4
u = Re cos 7t − ϕ . (3.63)
4
To√satisfy the initial conditions, we need
p R cos ϕ = 1, and R sin√
ϕ=
9/ 7. From this we get that R = 2 22/7, and ϕ = arctan(9/ 7).

Question 2: Sketch the solution.

Answer: From (3.63), we know that the solution oscillates between
Re−t/4 and Re−t/4 . These are the red dashed curves in Figure 3.6
(the under-damped plot). Since u′ (0) > 0, then the solution starts
out at u(0) = 1, and moves upward. From that point on it simply
bounces back and forth between the red dashed curves.

Question 3: When is the first time that u(t) = 0?

Answer: Writing
√ the solution as u = Re−t/4 cos θ, the phase func-
1
tion is θ = 4 7t − ϕ. Sketching this as in Figure 3.5, but now with
3.10. Linear Oscillator 69

0
3

0
3

-3
3

-3
0 2 4 6 8 10 12 14 16 18 20

Figure 3.6. Response of a damped mass-spring system, depending on the

strength of damping that is present. The dashed red curves in the two lower graphs are
the functions ±Reλt , where λ = −c/(2m).

0 < ϕ < π/2, it is seen that the first

√ time
√ cos θ = 0 is when θ = π/2.
Consequently, t = 2[π + 2arctan( 7)]/ 7. 

Example 3: For a given mass-spring-dashpot system, how does the so-

lution change as the damping coefficient changes?
√

Answer: Taking m = 2, k = 1, then, from (3.58), r = − c ± c2 − 8 /4.
Using the initial conditions u(0) = 1, and u′ (0) = 2, the resulting solu-
tion is shown in Figure 3.6, for different values for the damping constant.
The values
√ used give rise to: over-damping (c = 6), critically damped
(c = 2), under-damped (c = 1), and weakly damped (c = 1/40). To
say it is weakly damped means that it is under-damped, and c is so
small that the solution resembles the periodic solution of an undamped
oscillator, at least at the beginning. Eventually, the damping does reduce
the amplitude enough to be noticeable.
For over-damping, and critical damping, except near the beginning,
the solution simply decays monotonically to zero. In comparison, for both
under-damped cases the solution oscillates as it decays to zero. In both
cases the solution bounces back and forth between the two dashed red
curves, which are just the functions Reλt and −Reλt . 
70 Chapter 3. Second-Order Linear Equations

3.10.4 Resonance
We now consider what happens when a simple harmonic oscillator is
forced periodically. The specific equation to solve is

mu′′ + ku = F cos ωt, (3.64)

where ω is the driving frequency, and F is the amplitude of the forcing

(both ω and F are positive). Assuming u(0) = u′ (0) = 0, the resulting
solution is
F
  

 cos(ωt) − cos(ω 0 t) , if ω 6= ω0 ,
k − mω 2

u= (3.65)
F
 √ t sin(ω0 t), if ω = ω0 ,


2 km
where ω0 is given in (3.50).
What is of interest here is that, when the system is driven at its natu-
ral frequency ω0 , the solution is an oscillatory function whose amplitude
becomes unbounded as t increases. This happens even though the am-
plitude of the forcing is constant. This is an example of what is called
resonance. An example of a resonant solution is shown in Figure 3.7,
upper.
Resonance is a particularly important phenomena in science and engi-
neering, and it is often something that is to be avoided. As an example, a
wing on an airplane can go into a flapping motion. This can be modeled
as a simple harmonic oscillator, and under certain conditions the wing
can start to go into resonance. This is known as flutter, and the resulting
large oscillations can lead to the wing breaking off (which can be upsetting

50
u(t)

-50
0 50 100 150 200 250 300 350 400

10
u(t)

-10
0 50 100 150 200 250 300 350 400
t

Figure 3.7. Upper: Resonant solution given in (3.65) when ω√= ω0 = 1/3,
F = 1, m = 9 and k = 1. The red dashed lines correspond to ±F t/2 km. Lower:
Solution when a dashpot, with c = 1/4, is included.
3.10. Linear Oscillator 71

to those in the airplane). What is a concern is that this will happen no

matter what the value of F , as long as it’s nonzero. So, even a very small
force, what would normally be considered to be inconsequential, can lead
to extremely large oscillations.
One way to avoid resonance is to include a damping mechanism in the
system. With the dashpot we introduced earlier, the equation to solve is

mu′′ + cu′ + ku = F cos ωt. (3.66)

The forcing no longer contains a solution of the associated homogeneous

equation, and so resonance will not occur. However, it is often the case
that the damping is weak. This means that if ω = ω0 , then the solution
will start out like it’s going into resonance, but eventually the damping
will stop this. An example of what happens is shown in Figure 3.7, lower.
This brings us to the question to be considered. Using the flutter
example, the question is: we don’t want the wings to break off, so just
how large do the oscillations get before the damping stops this? To an-
swer this, it is the particular solution that is responsible for the growing
oscillations. So, for the weakly damped case we are considering, only the
particular solution is considered. To find the particular solution of (3.66),
the assumption that u = A cos ωt + B sin ωt leads to the requirement that
A and B satisfy

m(ω02 − ω 2 )A + cωB = F,
−cωA + m(ω02 − ω 2 )B = 0.

From this one obtains

m(ω02 − ω 2 ) cω
A= F, B= F.
c2 ω 2 + m2 (ω02 − ω 2 )2 c2 ω 2 + m2 (ω02 − ω 2 )2

Now, to determine the amplitude of the resulting oscillation, it makes

things easier to write the solution in the form u = R cos(ωt − ϕ). This
requires that R cos ϕ = A and R sin ϕ = B, and therefore
p 1
R= A2 + B 2 = p F. (3.67)
c2 ω 2 + m2 (ω02 − ω 2 )2

The amplitude R is plotted in Figure 3.8 as a function of the driving

frequency ω in the particular case of when F = 1, m = 1, and k = 1/2.
What is seen is that the smaller the damping coefficient c, the more peaked
the response becomes. Also, the peak response occurs at a frequency
smaller than the natural frequency ω0 , but this difference decreases as c
is reduced.
Our flutter question is answered by determining what driving fre-
quency ω gives the largest value for R. Taking the derivative of R with
72 Chapter 3. Second-Order Linear Equations

Figure 3.8. The amplitude (3.67) of the forced, but damped, oscillator, as a
function of the driving frequency. Note that ω0 is the natural frequency of the undamped
oscillator.

respect to ω, and setting it to zero gives us that ω = ωM , where

q
ωM = ω02 − ωc2 (3.68)
√
and ωc = c/( 2m). The resulting maximum for R is therefore
1
RM = q F. (3.69)
c ω02 − 12 ωc2

Now, suppose that for the flutter problem it is found experimentally that
the wings won’t break off if the amplitude of the oscillation satisfies R ≤
Rb . Based on our calculations, this means that the damping coefficient c
must be large enough that RM ≤ Rb .
Reality Check: The resonance phenomena considered here is not possi-
ble for the mass-spring system envisioned in Figure 3.2. As the oscillations
grow, as predicted by (3.65), they will eventually get to the point that the
mass will start banging up against the upper support. Presumably, as the
amplitude grows, our simple linear model is no longer valid, and a more
physically realistic, nonlinear, model is necessary. Even so, the simple
model is useful as it provides information about the onset of resonance.
Units and Values: In the exercises, the value to use for g is usually
stated. If it is not given, then you should leave g unevaluated. Whatever
value is used, it is only approximate. If a more physically realistic value is
needed, then you should probably use the Somigliana equation. Finally,
weight is a force, so for an object that weighs w lbs, its mass can be
determined from the equation w = mg.

Exercises
In answering
√ the following questions, do not numerically evaluate numbers
such as 2, π/3, e2 , ln(4/3), etc. The exception to this is when the
Exercises 73

question explicitly asks you to compute the answer.

1. Write the following in the form u = R cos(ω0 t − ϕ).
√
a) u = cos 3t + sin 3t c) u = − 13 3 cos t + sin t
√
b) u = 3 cos πt − sin πt d) u = −4 cos 2t − 4 sin 2t

2. A block weighing 2 lb stretches a spring 6 in. Assume that the mass

is pulled down an additional 3 in and then released from rest. Assume
that g = 32 ft/s2 .
a) What IVP does u(t) satisfy?
b) What is the solution of the IVP?
c) What is the natural frequency, period, and amplitude of the motion?
d) Sketch the solution for 0 ≤ t ≤ 3T .
e) Is the restoring force in the spring ever zero? What is the minimum
value of the force in the spring?
3. A mass of 100 gm stretches a spring 56 m. Assume that the mass is
pulled down a distance of 1 m, and then set in motion with an upward
velocity of 2 m/s. Assume that g = 10 m/s2 .
a) What IVP does u(t) satisfy?
b) What is the solution of the IVP?
c) What is the natural frequency, period, and amplitude of the motion?
d) When does the mass first return to its steady state position?
e) Sketch the solution for 0 ≤ t ≤ 3T .
f) What is the first time the force F is zero?
4. A mass of 1 kg stretches a spring 10 cm. Assume that the mass is
pushed upward a distance of 5 cm, and then set in motion with a
downward velocity of 50 cm/s. Assume that g = 10 m/s2 .
a) What IVP does u(t) satisfy?
b) What is the solution of the IVP?
c) What is the natural frequency, period, and amplitude of the motion?
d) Sketch the solution for 0 ≤ t ≤ 3T .
e) What is the largest value of the restoring force in the spring? When
is the first time it equals this value?
5. According to Archimedes’ principle, an object that is completely or
partially submerged in water is acted on by an upward (buoyant) force
equal to the weight of the displaced water. You are to use this for
the following situation: A cubic block of wood, with side l and mass
density ρ, is floating in water. If the block is slightly depressed and then
released, it oscillates in the vertical direction. Derive the differential
equation of motion and determine the period of the motion. In doing
this let ρ0 be the mass density of the water, and assume that ρ0 > ρ.
74 Chapter 3. Second-Order Linear Equations

6. In a mass-spring system, suppose the mass is pulled down a distance

d and released from rest.
a) If the resulting natural frequency is 10 s−1 when d = 0.1 m, what is
the natural frequency when d = 0.2 m?
b) Suppose the amplitude of the motion is 0.1 m. Is it possible to
change the initial velocity, keeping d unchanged, so the amplitude
of the motion is 0.2 m?
c) Suppose the natural frequency is 10 s−1 when d = 0.1 m. Is it
possible to change the initial velocity, keeping d unchanged, so the
natural frequency is 20 s−1 ?

Damping

7. A block weighing 16 lb stretches a spring 6 in. The mass is attached

to a viscous damper with a damping constant of 2 lbs·s/ft. Assume
that the mass is set in motion from its equilibrium position with a
downward velocity of 4 in/s. Also, assume that g = 32 ft/s2 .
a) What IVP does u(t) satisfy?
b) What is the solution of the IVP?
√
c) Sketch the solution for 0 ≤ t ≤ π 15/5.
d) What is the largest value of u(t)?
8. Suppose you construct a mass-spring-dashpot system as shown in Fig-
ure 3.1. In this problem assume that g = 10 m/s2 .
1
a) If the spring is stretched 10 m by a force of 12 N, what is the spring
constant?
b) If the dashpot exerts a force of −3 N when the velocity is 1 m/s,
what is the damping constant?
c) Suppose the mass is 21 kg, and it is pulled up 1 m from its rest
position and given an initial downward velocity of 2 m/s. What
IVP does u(t) satisfy?
d) What is the solution of the IVP?
e) Sketch the solution for 0 ≤ t ≤ 10π.
9. The general solution for the under-damped case is given in (3.62).
Suppose the initial conditions are u(0) = u0 and u′ (0) = u′0 .
a) Show that
s 2
u′0 − λu0

R= u20 + .
µ

b) How does Figure 3.3 change?

c) What are R and ϕ if u′0 = 0?
Exercises 75

10. It is often stated that “the key difference between critical damping and
overdamping is that critical damping provides the quickest approach
to zero amplitude.” However, this statement is not true. This problem
investigates this for the case of when m = 1, k = 4, u(0) = 1, and
u′ (0) = −4.
a) Find the solution when c = 5, which is the over-damped case, and
when c = 4, which is the critically damped case. Sketch both
solutions on the same axes. Explain why the statement is not true.
b) Solve the two problems in part (a) but use the general initial condi-
tions u(0) = u0 and u′ (0) = u′0 . Use this to explain how to modify
the statement so that it is true.
11. It is usually stated that negative damping is unstable. For the mass-
spring-dashpot system, negative damping means that c is negative.
From the solution, explain why the system is unstable for any nonzero
initial conditions.
Resonance and Forced Motion
12. A block weighing 4 lb stretches a spring 1.5 in. Assume that the block
is acted on by a periodic forcing as in (3.64), with F = 3 lb and ω =
16 /sec. At the start, the block is not moving and it is at its rest
position. Assume that g = 32 ft/s2 .
a) What IVP does u(t) satisfy?
b) What is the solution of the IVP?
c) Sketch the solution for 0 ≤ t ≤ π.
13. Suppose that a spring-mass system is at rest but, starting at t = 0,
the mass is subjected to a force of 5 cos 3t N. Assume that the mass is
2 kg, and the spring constant is 18 kg/s2 .
a) What IVP does u(t) satisfy?
b) What is the solution of the IVP?
c) Sketch the solution for 0 ≤ t ≤ 4π.
14. Suppose the forcing in (3.66) is replaced with F sin ωt. Does this
change (3.67)?
15. This exercise considers what happens when the forcing in (3.66) con-
sists of a combination of driving frequencies.
a) Suppose the forcing is

F0 cos ω0 t + F1 cos ω1 t + F2 cos ω2 t,

where the Fi ’s are nonzero, and the ωi ’s are all different, with ω0
given in (3.50). Does resonance still occur?
b) Suppose the forcing is F0 cos ω0 t cos ω1 t, where F0 is nonzero, ω1 6=
ω0 , and ω0 given in (3.50). Does resonance still occur?
76 Chapter 3. Second-Order Linear Equations

3.11 Euler Equation

Although second-order equations with constant coefficients are the ones
that most often arise in applications, there is a notable exception to this
statement. This is the Euler equation, which is
d2 y dy
x2 2
+ bx + cy = 0, (3.70)
dx dx
where b and c are constants. The reason this equation arises as often as
it does is that it comes from using polar coordinates when solving what
is known as Laplace’s equation (see Section 7.8.2).
A complication that arises with (3.70) is that it is not a second-order
differential equation when x = 0. For this reason, x = 0 is referred to
as a singular point for the equation. This is an issue as it is often the
case that the interval used when solving Euler’s equation has the form
0 ≤ x < L. What condition, if any, you can impose at x = 0 is a question
we will consider below.
In what follows it is assumed that x > 0. Solving (3.70) is rather easy,
as one just assumes a solution of the form y = xr . Since y ′ = rxr−1 and
y ′′ = r(r − 1)xr−2 , then from (3.70 ) we get that
r(r − 1) + br + c = 0, (3.71)
The solutions of this quadratic equation are
1 p 
r= 1 − b ± (1 − b)2 − 4c . (3.72)
2
Just as in Section 3.3, what happens next depends on the values of r
obtained from this solution.

Two Real Roots

When there are two real-valued roots, say, r1 and r2 , then the two cor-
responding solutions of (3.14) are y1 = xr1 and y2 = xr2 . It is left as an
exercise to show they are independent. Therefore, the resulting general
solution of (3.14) is
y = c 1 x r1 + c 2 x r2 , (3.73)
where c1 and c2 are arbitrary constants.

One Real Root

When there is only one root, then you use reduction of order. This means
that to find a second solution, assume that y = w(x)xr . Proceeding as
in Section 3.3.2, one finds that w = ln x. Therefore, the resulting general
solution of (3.14) is
y = c1 xr + c2 ln(x)xr ,
where c1 and c2 are arbitrary constants.
Exercises 77

Complex Roots
In this case, the roots can be written as r = λ ± iµ, where
1
λ = (1 − b), (3.74)
2
and
1p
µ= 4c − (1 − b)2 . (3.75)
2
It is assumed here that 4c > (1 − b)2 . Writing the general solution as
in (3.73), and then separating into real and complex parts using Euler’s
formula, one finds that the resulting general solution can be written as

y = d1 xλ cos(µ ln x) + d2 xλ sin(µ ln x), (3.76)

where d1 and d2 are arbitrary constants.

3.11.1 Examples

Example 1: Find the solution of x2 y ′′ + 2xy ′ − 6y = 0, for 0 < x < 2,

that is bounded for 0 < x < 2 and satisfies y(2) = 1.
Answer: Substituting in y = xr , one gets the equation r2 +r−6 = 0.
The solutions of this are r = −3, and r = 2. So, the general solution
of the differential equation is

y = c1 x−3 + c2 x2 .

The requirement that y is bounded means that c1 = 0. As for

y(2) = 1, we need c2 = 1/4. Therefore, the solution is
1
y(x) = x2 . 
4

Example 2: Find the general solution of 4x2 y ′′ +17y = 0, for 0 < x < ∞.
Answer: Substituting in y = xr , one gets the equation 4r2 − 4r +
17 = 0. The solutions of this are r = 21 ± 2i. So, from (3.76), the
general solution is
√ √
y = d1 x cos(2 ln x) + d2 x sin(2 ln x). 

Exercises
1. Assuming x > 0, find the general solution of the following Euler equa-
tions.
78 Chapter 3. Second-Order Linear Equations

a) x2 y ′′ − 3xy ′ + 4y = 0 f) 5x2 y ′′ + 12xy ′ + 2y = 0

b) x2 y ′′ − 5xy ′ + 10y = 0 g) x2 y ′′ + xy ′ = 0
c) 6x2 y ′′ + 7xy ′ − y = 0 h) x2 y ′′ − 2xy ′ = 0
d) x2 y ′′ + y = 0 i) x2 y ′′ − xy ′ − n(n + 2)y = 0,
e) x2 y ′′ − 3xy ′ + 13y = 0 where n is a positive integer

2. Find the solution of the following problems. Before doing these prob-
lems, you might want to review Exercise 3, on page 63.
a) x2 y ′′ − 2xy ′ + 2y = x3 ex , where y(1) = 0, and y ′ (1) = 0
b) x2 y ′′ − 4xy ′ + 4y = −2x2 + 1, where y(1) = 0, and y ′ (1) = 0
c) x2 y ′′ − xy ′ + y = ln x, where y(1) = 0, and y ′ (1) = 0
d) xy ′′ + y ′ = x, where y(1) = 1, and y ′ (1) = −1
e) (x − 1)2 y ′′ + (x − 1)y ′ − y = 0, where y(2) = 1, and y ′ (2) = 0

3.12 Guessing the Title of the Next Chapter

Since Chapter 2 is about first-order equations, and this chapter is about
second-order equations, you might expect the next chapter to be about
third-order equations. This was often how older textbooks were written,
where the next chapter would be titled Higher-Order Equations, or some-
thing similar. Although it is possible to find applications that involve
higher-order equations, such as viscoelasticity, they are not that com-
mon. Moreover, the usual method for solving higher-order equations is
to first rewrite them in system form. This is certainly the approach used
when solving them numerically. Well, as it turns out, the next chapter is
about linear systems and the following chapter is on nonlinear systems.
So, although third-order and higher equations are not considered in this
text, the methods used to solve them are.
Chapter 4

Linear Systems

This chapter, and the one that follows, consider problems that involve
two or more first-order ordinary differential equations. Together the equa-
tions form what is called a first-order system. These are very common.
To explain why, it is worth considering a couple of examples.

Example 1: Mechanics
As stated on several occasions earlier in this text, one of the biggest
generators of differential equations is Newton’s second law, which states
that F = ma. To demonstrate its connection with a system of differential
equations, let x(t) denote the position of an object. The velocity is then
v = x′ (t), and the acceleration is a = x′′ (t). So, F = ma can be written
as mv ′ = F . Along with the equation x′ = v, the resulting system is

dx
= v,
dt
dv 1
= F.
dt m
As an example, for a uniform gravitation field, and including air resis-
tance, then F = −mg − cv (see Section 2.3.2). In this case, the system
becomes

x′ = v,
c
v ′ = −g − v.
m
This is a linear first-order system for x and v. It is also inhomogeneous
since x ≡ 0 and v ≡ 0 is not a solution. 
Introduction to Differential Equations, M. H. Holmes, 2020

79
80 Chapter 4. Linear Systems

Example 2: Epidemics
Epidemics, such as the black death and cholera, have come and gone
throughout human history. Given the catastrophic nature of these events
there is a long history of scientific study trying to predict how and why
they occur. One of particular prominence is the Kermack-McKendrick
model for epidemics. This assumes the population can be separated into
three groups. One is the population S(t) of those susceptible to the
disease, another is the population I(t) that is ill, and the third is the
population R(t) of individuals that have recovered. A model that accounts
for the susceptible group getting sick, the subsequent increase in the ill
population, and the eventual increase in the recovered population is the
following set of equations [Holmes, 2019]
dS
= −k1 SI,
dt
dI
= −k2 I + k1 SI,
dt
dR
= k2 I.
dt
Given the three groups, and the letters used to designate them, this is an
example of what is known as a SIR model in mathematical epidemiology.
For us, this is an example of a nonlinear first-order system for S, I, and
R. The reason it is nonlinear is the SI term that appears in the first two
equations.

As you might expect, solving a nonlinear system can be challenging.

So, in this chapter, we will concentrate on linear systems. In the next
chapter, nonlinear problems are considered.

4.1 Linear Systems

To get things started, consider the problem of solving
x′ = ax + by, (4.1)
′
y = cx + dy. (4.2)
This is a first-order, linear, homogeneous system. In these equations, x(t)
and y(t) are the dependent variables, and a, b, c, and d are constants.
This can be written in system form as
! ! !
d x a b x
= .
dt y c d y
A simpler way to write this is as
d
x = Ax, (4.3)
dt
4.1. Linear Systems 81

where the vector is !

x
x= ,
y
and the matrix is !
a b
A= . (4.4)
c d
The equation in (4.3) plays a central role throughout this chapter. Written
in this way, we could be dealing with 20 equations, or 200 equations, and
not just the two in (4.1) and (4.2).
For those a bit rusty on the basic rules for working with matrices and
vectors, a short summary is provided in Appendix A.
Before getting into the discussion of how to solve (4.3), it is worth
considering what we already know about the solution.

4.1.1 Example: Transforming to System Form

In Section 3.5, Example 1, we found that for

y ′′ + 2y ′ − 3y = 0 (4.5)

the roots of the characteristic equation are r1 = −3 and r2 = 1. The

resulting independent solutions are y1 = e−3t and y2 = et . In this exam-
ple, the differential equation, along with its solutions, are translated into
vector form.

a) Write (4.5) as a linear first-order system as in (4.3).

The standard way to do this is to let v = y ′ , so the differential equation
can be written as v ′ + 2v − 3y = 0, or equivalently, v ′ = 3y − 2v. This,
along with the equation y ′ = v, gives us the system

y ′ = v,
v ′ = 3y − 2v.

In other words, we have an equation of the form (4.3), where

! !
y 0 1
x= , and A= .
v 3 −2

b) Write the two linearly independent solutions in vector form.

For y1 = e−3t , then v1 = y1′ = −3e−3t . Letting x1 be the solution vector
coming from y1 , then
! ! !
y1 e−3t 1
x1 = = = e−3t = a1 er1 t ,
v1 −3e−3t −3
82 Chapter 4. Linear Systems

where r1 = −3 and !
1
a1 = .
−3

Similarly, since v2 = y2′ = et , then letting x2 be the vector version of y2 ,

! ! !
y2 et 1 t
x2 = = = e = a2 e r 2 t ,
v2 et 1

where r2 = 1 and !
1
a2 = .
1

c) Write the general solution in vector form.

The general solution for the second-order equation is y = c1 y1 + c2 y2 .
From this, we get that v = y ′ = c1 y1′ + c2 y2′ . Therefore, the general
solution vector is
! ! ! !
y c 1 y1 + c 2 y2 c 1 y1 c 2 y2
x= = = +
v c1 y1′ + c2 y2′ c1 y1′ c2 y2′
= c1 x1 + c2 x2 .  (4.6)

A very useful observation to make about the above example is that

the linearly independent solutions have the form x = aert , where a is
a constant vector. In fact, when the time comes to solve (4.3) we will
simply assume that x = aert , and then find r and a. Also, note that
for the single linear equation x′ = ax, there is one linearly independent
solution. As the above example shows, for two linear first-order equations
there are two linearly independent solutions. Consequently, it should not
be a surprise to find out that for n linear first-order equations there are
n linearly independent solutions.

4.1.2 General Version

We are going to consider solving homogeneous linear first-order systems.
Assuming there are n dependent variables, then the system can be written
as

x′1 = a11 x1 + a12 x2 + · · · a1n xn

x′2 = a21 x1 + a22 x2 + · · · a2n xn
.. .. ..
. . .
x′n = an1 x1 + an2 x2 + · · · ann xn ,
Exercises 83

where the aij ’s are constants. This can be written as

d
x = Ax, (4.7)
dt
where A is an n × n matrix, and x is an n-vector, given, respectively, as

x1
   
a11 a12 ··· a1n
 a21 x 
a22 ··· a2n   2
A= . .. ..  and x=
 ..  .
  
 .. . ··· .   . 
an1 an2 ··· ann xn

For an initial value problem, an n-vector x0 would be given, and the

condition to be satisfied would be x(0) = x0 .
Because (4.7) is linear and homogeneous, the principle of superposition
holds (see page 5). Therefore, if x1 and x2 are solutions of (4.7), then

x = c1 x 1 + c2 x 2

is a solution for any values of the constants c1 and c2 .

d
As a final comment, the inhomogeneous equation dt x = Ax + f is not
considered in this chapter, but it is considered in Section 6.6.

Exercises
1. Write the following as x′ = Ax, making sure to identify the entries in
x and A. If initial conditions are given, write them as x(0) = x0 .

a) u′ = u − v d) u′ = u − v
v ′ = 2u − 3v v ′ = 2u − 3v
b) 2u′ = −u u(0) = −1, v(0) = 0
3v ′ = u + v e) u′ = 2u − w
c) u′ = u − v + 2w v′ = u + v + w
v′ = u 3w′ = 2v + 6w
w′ = −u + 5v u(0) = −1, v(0) = 0, w(0) = 3

2. For the following: a) Write the equation in the form x′ = Ax. b) Find
the general solution of the second-order equation and then write it in
vector form as x = c1 x1 + c2 x2 , where x1 = a1 er1 t and x2 = a2 er2 t .
Make sure to identify a1 , a2 , r1 and r2 .

a) y ′′ + 2y ′ − 3y = 0 c) 3y ′′ + 4y ′ + 3y = 0
b) 4y ′′ + 3y ′ − y = 0 d) y ′′ + 4y ′ = 0
84 Chapter 4. Linear Systems

3. Show that the given vector x is a solution of the differential equation.

Also, what initial condition does x satisfy?
a) ! !
1 2 4
′
x = x, x= e2t
2 −2 2

b) ! !
3 1 1
x′ = x, x=3 et
2 2 −2

c) ! ! !
′
2 0 1 2t 0
x = x, x= e + e−3t
0 −3 0 2

d)
1
! !
2 1 cos t
′
x = 1
x, x= et/2
−1 2
− sin t

4. This problem considers other possibilities for transforming a second-

order equation into a first-order system.
a) Assuming that c 6= 0, show that (4.1), (4.2) can be reduced to the
second-order linear equation

y ′′ − (a + d)y ′ + (ad − bc)y = 0.

b) Using the result from part (a), transform y ′′ + 2y ′ − 3y = 0 into a

first-order system where none of the entries in A are zero.

4.2 General Solution of a Homogeneous Equation

The problem considered here is

d
x = Ax, for t > 0. (4.8)
dt
From (4.6), as well as Exercise 2 in the previous section, we have an idea
of what the general solution of this equation looks like. Namely, if we are
able to find n linearly independent solutions x1 (t), x2 (t), . . ., xn (t), then
the general solution can be written as

x(t) = c1 x1 (t) + c2 x2 (t) + · · · + cn xn (t), (4.9)

where c1 , c2 , . . ., cn are arbitrary constants.

The requirement to be linearly independent is a simple generalization
of the definition given in Section 3.2. Namely, x1 (t), x2 (t), . . ., xn (t) are
4.3. Review of Eigenvalue Problems 85

linearly independent if, and only if, the only constants c1 , c2 , . . ., cn

that satisfy
c1 x1 + c2 x2 + · · · + cn xn = 0, ∀ t ≥ 0, (4.10)
are c1 = 0, c2 = 0, . . ., cn = 0. In the above equation, 0 is the zero
vector, which means that all of its components are zero. Also, the symbol
∀ is a mathematical shorthand for “for all” or “for every.”
In the last chapter the Wronskian was used to determine indepen-
dence. It is possible to also use the Wronskian with (4.8), but this is not
particularly useful for larger n. There is an easier way to show indepen-
dence, and this will be explained in Section 4.4.
The general solution of (4.8) is found by assuming that x = aert ,
where a is a constant vector. Differentiating this expression, x′ = raert ,
and so (4.8) becomes raert = A(aert ). Since ert is never zero we can
divide by it, which gives us the equation
Aa = ra. (4.11)
What we want are nonzero solutions of this equation, and so we require
that a 6= 0. This problem for r and a is called an eigenvalue problem,
where r is an eigenvalue, and a is an associated eigenvector. This
is one of the core topics covered in linear algebra. We do not need to
know the more theoretical aspects of this problem, but we certainly need
to know how to solve it. So, for completeness, the more pertinent aspects
of an eigenvalue problem are reviewed next.
It is worth pointing out that it is possible to solve (4.8) without using
eigenvalues and eigenvectors, and how this is done is explained in Section
6.6.

4.3 Review of Eigenvalue Problems

Given an n×n matrix A, its eigenvalues r and the associated eigenvectors
a are found by solving
Aa = ra. (4.12)
It is required that a is not the zero vector. There are no conditions placed
on r, and it can be real or complex valued.
In preparation for solving the above equation, it is first rewritten as
Aa − ra = 0, or equivalently as
(A − rI)a = 0. (4.13)
The n × n matrix I is known as the identity matrix and it is defined as
 
1 0 ··· 0
0 1 ··· 0
I ≡ . .. ..  .
 
 .. . ··· .
0 0 ··· 1
86 Chapter 4. Linear Systems

For example, when n = 2 and n = 3,

 
! 1 0 0
1 0
I= and I = 0 1 0 .
 
0 1
0 0 1

In linear algebra it is shown that for the equation (4.13) to have a

nonzero solution, it is necessary that the matrix A − rI be singular, or
non-invertible. What this means is that the determinant of this matrix
is zero. This gives rise to the following method for solving the eigenvalue
problem.

Eigenvalue Algorithm. The procedure used to solve the eigenvalue prob-

lem consists of two steps:

1. Find the r’s by solving

det(A − rI) = 0. (4.14)

This is known as the characteristic equation, and the left-hand-

side of this equation is an nth degree polynomial in r.

2. For each eigenvalue r, find the associated eigenvectors by finding the

nonzero solutions of
(A − rI)a = 0. (4.15)

In this textbook we are mostly interested in systems involving two equa-

tions. For those who might not remember, the determinant of a 2 × 2
matrix is defined as
!
a11 a12
det ≡ a11 a22 − a12 a21 .
a21 a22

In the second step of the algorithm, when solving (4.15), we are inter-
ested in finding the vectors that can be used to form the general solution
of this equation. To say this more mathematically, we want to find lin-
early independent solutions. For those you might not remember what this
is, the definition is given next.

Linearly Independent. The vectors a1 , a2 , · · · , ak are linearly inde-

pendent if, and only if, the only numbers c1 , c2 , . . ., ck that satisfy

c1 a1 + c2 a2 + · · · + ck ak = 0, (4.16)

are c1 = c2 = · · · = ck = 0. If it is possible for any of the ci ’s to be

nonzero, then the vectors are linearly dependent.
4.3. Review of Eigenvalue Problems 87

In n dimensions, it is not possible to have more than n linearly inde-

pendent vectors. Consequently, n is the maximum number of linearly
independent eigenvectors you can find for an n × n matrix A. Finally,
as will be seen in one of the examples to follow, eigenvectors can contain
complex numbers. When this happens, the ci ’s in the above equation
must be allowed to be complex-valued.
The following examples all involve 2 × 2 matrices. What is illustrated
are the various situations that can arise with eigenvalue problems. In
these examples, the eigenvector will be written in component form as
!
a
a= . (4.17)
b

Example 1: Two Real Eigenvalues

For !
2 1
A= ,
1 2

we get that
! ! !
2 1 1 0 2−r 1
A − rI = −r = .
1 2 0 1 1 2−r

Since det(A − rI) = (2 − r)2 − 1 = r2 − 4r + 3, then the characteristic

equation (4.14) is r2 − 4r + 3 = 0. Solving this we get that the eigenvalues
are r1 = 3 and r2 = 1. For r1 , (4.15) takes the form
! ! !
−1 1 a 0
= .
1 −1 b 0

In component form, we have that

−a + b = 0,
a − b = 0.

The solution is b = a, and so the eigenvectors are

! !
a a
a= = = a a1 , (4.18)
b a

where !
1
a1 = . (4.19)
1
88 Chapter 4. Linear Systems

For the second eigenvalue r2 = 1, one finds that the eigenvectors have the
form a = aa2 , where a is an arbitrary nonzero constant and
!
1
a2 = .
−1

The eigenvectors a1 and a2 are independent. To show this, note that

! ! !
1 1 c1 + c2
c1 a1 + c2 a2 = c1 + c2 = .
1 −1 c1 − c2

So, from (4.16), if c1 a1 + c2 a2 = 0, then c1 + c2 = 0 and c1 − c2 = 0. From

the last equation, c1 = c2 , and inserting this into the first equation yields
2c2 = 0. So, c2 = 0, and this also means that c1 = 0. Therefore, a1 and
a2 are independent. 

There is an important observation that needs to be made here. In

the above example, it was shown that eigenvectors for different eigenval-
ues are linearly independent. This is always true, and this is important
enough that it needs to be stated more prominently.

Different Eigenvalues Test. If a1 , a2 , · · · , ak are eigenvectors corre-

sponding to different eigenvalues for a matrix A, then these vectors are
linearly independent.

The above test applies irrespective of whether the eigenvalues are real or
complex valued. It also is not limited to a 2 × 2 matrix, and holds in the
general case of when the matrix is n × n.

Example 2: One Eigenvalue But Two Independent Eigenvectors

When !
3 0
A= ,
0 3
the characteristic equation is (r − 3)2 = 0. So the only eigenvalue is
r1 = 3. In this case, !
0 0
A − rI = . (4.20)
0 0
This means that all vectors are solutions of (4.15). In other words, the
solutions are
! ! !
a a 0
a= = + = a a1 + b a2 ,
b 0 b
4.3. Review of Eigenvalue Problems 89

where ! !
1 0
a1 = and a2 = . (4.21)
0 1
To check on independence, we are not able to use the Different Eigen-
values Test given above because a1 and a2 are eigenvectors for the same
eigenvalue. To use the definition, note that
! ! !
1 0 c1
c1 a1 + c2 a2 = c1 + c2 = .
0 1 c2

So, if c1 a1 + c2 a2 = 0, then we conclude that c1 = c2 = 0. Therefore, a1

and a2 are independent. 

Example 3: Complex-Valued Eigenvalues

For the matrix !
1 2
A= ,
− 12 1

the characteristic equation is r2 − 2r + 2 = 0. The resulting eigenvalues

are r = 1 + i and r = 1 − i. Proceeding as usual, for r1 ,
!
i 2
A − r1 I = .
− 12 i

This means that (4.15) requires that −ia + 2b = 0, or equivalently, a =

−2ib. So, the eigenvectors are
! !
a −2i
a= =b = b a1 ,
b 1

where !
−2i
a1 = .
1
Similarly, for r2 = 1 − i, one finds that the eigenvectors are

a = b a2 ,

where !
2i
a2 = .
1
Finally, because a1 and a2 are eigenvectors for different eigenvalues, they
are independent. 
90 Chapter 4. Linear Systems

There is an observation that needs to be made here. In the above

example, the eigenvalues have the form r1 = λ+iµ and r2 = λ−iµ, where
λ and µ are real numbers, with µ 6= 0. Because of this, the eigenvalues
are said to be complex conjugates. When a matrix only contains real
numbers, as in the last example, and it has complex eigenvalues, they
must occur as complex conjugates. Moreover, you should notice that
the respective eigenvectors a1 and a2 are also complex conjugates (if you
change i to −i in a1 , you get a2 ). This is useful information as it means
that once you know a1 , you immediately know a2 .

Example 4: Only One Independent Eigenvector

The matrix !
3 1
A= ,
0 3

has one eigenvalue r = 3 (similar to Example 2). In this case (4.15)

becomes !
0 3
A − rI = . (4.22)
0 0

This means that b = 0. Consequently, the eigenvectors have the form

! !
a 1
a= =a = a a1 ,
0 0

where !
1
a1 = .
0

In the previous three examples involving 2 × 2 matrices we found two

linearly independent eigenvectors. This matrix is different as there is only
one. An n × n matrix that has fewer than n independent eigenvectors is
said to be defective. So, the matrix of this example is defective, while
the matrices for the three previous examples are not defective. 

A semi-useful observation can be made here. The only 2 × 2 matrices

that have only one eigenvalue, and are not defective, have the form A =
aI. In this case, the eigenvalue is r = a and the eigenvectors are given in
(4.21). This is what happened in Example 2, where a = 3. So, if you have
a 2 × 2 matrix with only one eigenvalue, and A 6= aI, then the matrix is
defective. This is what happened in Example 4.
Exercises 91

Exercises
1. Determine whether the following pairs of vectors are linearly indepen-
dent.
! ! ! !
1 2 2 −1
a) a1 = , a2 = c) a1 = , a2 =
2 1 −8 4
! ! ! !
1 −3 −5 1
b) a1 = , a2 = d) a1 = , a2 =
−1 3 10 2

2. The following matrices have two real-valued eigenvalues. Find the

eigenvalues, and two linearly independent eigenvectors.
! !
2 1 −2 −7
a) b)
4 −1 1 6

3. The following matrices have complex-valued eigenvalues. Find the

eigenvalues, and two linearly independent eigenvectors.
! !
2 −4 2 13
a) b)
1 2 −1 −4

4. Show that the following matrices are defective.

! !
3 −2 −1 1
a) b)
2 −1 −9 5

4.4 Solving a Homogeneous Equation

As stated earlier, given an n × n matrix A, to find the general solution of
d
x = Ax, (4.23)
dt
you start by assuming that x = aert , where a is a constant vector. Sub-
stituting this into the differential equation, and simplifying, leads to the
eigenvalue problem
Aa = ra. (4.24)
If A is not defective, then there are n linearly independent eigenvectors
a1 , a2 , . . ., an . Letting r1 , r2 , . . ., rn be their respective eigenvalues, then
the general solution of (4.23) can be written as
x = c1 a1 er1 t + c2 a2 er2 t + · · · + cn an ern t , (4.25)
92 Chapter 4. Linear Systems

where the ci ’s are arbitrary constants.

The vectors xj = aj erj t used in the above formula for the general
solution are linearly independent. The reason is that the test for inde-
pendence in (4.10) must hold at t = 0, and for t = 0 the equation reduces
to (4.16). Since the aj ’s are independent, it follows that the cj ’s are all
zero. Consequently, the vectors xj are linearly independent.
With the formula for the general solution in (4.25), all that is left to
do is consider how to rewrite it when the eigenvalues are complex and to
also determine what to do when the matrix is defective. A summary of
what follows in given in Section 4.5.

4.4.1 Complex-Valued Eigenvalues

As usual, when the roots are complex-valued there are options as to how
the general solution can be written. It is certainly possible to just use the
expression in (4.25). However, it is often easier to rewrite the solution so
as to avoid the use of complex variables. It is easiest to explain how this
is done using an example.

Example
The matrix in the differential equation,
!
1 2
x′ = x,
− 12 1

is the one considered in Example 3 of the previous section. The eigenval-

ues are r = 1 + i and r = 1 − i. Using the eigenvectors found earlier, the
general solution can be written as
! !
−2i (1+i)t 2i (1−i)t
x = c1 e + c2 e .
1 1

Because complex numbers are used for the r’s, both c1 and c2 must be
allowed to be complex-valued.
Given that x is real-valued, the coefficients c1 and c2 must be complex
conjugates. In other words, if c1 = α + iβ, where α and β are real-valued,
then it must be that c2 = α − iβ. We are going to separate the solution
into real and imaginary parts, which for the eigenvectors means that
! ! ! ! ! !
−2i 0 −2 2i 0 −2
= +i , and = −i .
1 1 0 1 1 0

It makes things a bit easier to write these as

! !
−2i 2i
= p + iq, and = p − iq,
1 1
4.4. Solving a Homogeneous Equation 93

where ! !
0 −2
p= , and q = .
1 0
Now, using Euler’s formula (3.14), we have that

x = (α + iβ)(p + iq)et (cos t + i sin t) + (α − iβ)(p − iq)et (cos t − i sin t)

= d1 (p cos t − q sin t)et + d2 (p sin t + q cos t)et

! !
2 sin t t −2 cos t t
= d1 e + d2 e,
cos t sin t

where d1 = 2α and d2 = −2β are arbitrary real-valued constants. 

General Formula
To summarize what was done in the above example, suppose that A is a
2 × 2 matrix with complex-valued eigenvalues r1 = λ + iµ and r2 = λ − iµ,
where λ and µ are real-valued with µ 6= 0. Also, assume that their
respective eigenvectors are a1 = p + iq and a2 = p − iq, where p and q
are vectors containing only real numbers. In this case, instead of writing
the general solution as

x = c1 a1 er1 t + c2 a2 er2 t ,

it can be written as

x(t) = d1 b1 eλt + d2 b2 eλt ,

where

b1 = p cos(µt) − q sin(µt),
b2 = p sin(µt) + q cos(µt),

and d1 and d2 are arbitrary real-valued constants. As a labor saving ob-

servation, it should be noted that p and q are known once the eigenvector
for r1 is found, which means you do not also need to find the eigenvector
for r2 .

4.4.2 Defective Matrix

The other case to consider is what to do when there are not enough
linearly independent eigenvectors, which means that A is defective. So,
suppose that A is a 2 × 2 matrix that has one eigenvalue r, and a is its
associated eigenvector. Based on the way we fixed the single root solution
in Chapter 3, you might expect for the vector version you should assume
94 Chapter 4. Linear Systems

a solution of the form x = btert . However, this does not work, and to
find a second independent solution, the assumption is that

x = atert + bert .

To find b, the above expression is substituted into the differential equation

to obtain
Ab = rb + a,
or equivalently
(A − rI)b = a. (4.26)
It is useful to know that we don’t need all solutions of this equation.
Rather, all we need is just one of them. Once this is determined, the
general solution is
x = c1 aert + c2 (ta + b)ert .

4.5 Summary for Solving a Homogeneous Equation

Assuming that A is 2 × 2, then the general solution of x′ = Ax is as given
below.

• When A is not defective.

– If A has real eigenvalues r1 and r2 , with respective eigenvectors

a1 and a2 , then
x = c1 a1 er1 t + c2 a2 er2 t . (4.27)
This expression can be used when r1 = r2 (in this case, just make
sure a1 and a2 are independent).

– If A has complex eigenvalues r = λ ± iµ (with µ 6= 0), with

respective eigenvectors p ± iq, then

x(t) = d1 b1 eλt + d2 b2 eλt , (4.28)

where

b1 = p cos(µt) − q sin(µt),
b2 = p sin(µt) + q cos(µt).

• When A is defective, with eigenvalue r and eigenvector a, then

x = c1 aert + c2 (ta + b)ert , (4.29)

where b is any solution of

(A − rI)b = a. (4.30)
4.5. Summary for Solving a Homogeneous Equation 95

Example 1 (real eigenvalues): Find the general solution of

!
′ 0 1
x = x.
2 1

Step 1: Find the eigenvalues and eigenvectors. Using the eigenvalue

algorithm, from (4.14),
!
−r 1
det(A − rI) = 0 ⇒ det =0
2 1−r
⇒ r2 − r − 2 = 0
⇒ r = −1, 2.

For r = −1, then from (4.15),

!
1 1
(A − rI)a = 0 ⇒ a=0 ⇒ a + b = 0.
2 2

So, b = −a, and this means that

! !
a a
a= = = a a1 ,
b −a

where !
1
a1 = .
−1

In a similar manner, one finds that for r = 2, an eigenvector is

!
1
a2 = .
2

Step 2: Since this is a non-defective matrix with real eigenvalues,

the general solution is
! !
1 1
x = c1 e−t + c2 e2t . 
−1 2

Example 2 (complex eigenvalues): Find the solution of the IVP:

! !
′ 2 1 0
x = x, where x(0) = .
−2 0 1
96 Chapter 4. Linear Systems

Step 1: Find the eigenvalues and eigenvectors. Using the eigenvalue

algorithm, from (4.14), you find that the eigenvalues are r1 = 1 + i
and r2 = 1 − i. To determine the eigenvector for r1 , we have that
! !
2 − (1 + i) 1 1−i 1
A − r1 I = = .
−2 −(1 + i) −2 −(1 + i)

So, writing a as in (4.17), then (A − r1 I)a = 0 can be written in

component form as

(1 − i)a + b = 0
−2a − (1 + i)b = 0.

Both equations lead to the conclusion that b = −(1 − i)a. So, the
eigenvectors are
! !
a 1
a= =a = a a1 .
b −1 + i

As explained earlier, it makes things easier to write a1 = p + iq,

which means that !
1
= p + iq,
−1 + i
where ! !
1 0
p= , and q= .
−1 1
Moreover, because the eigenvector for r2 = 1 − i is the complex
conjugate of a1 , then a2 = p − iq.
Step 2: Find the general solution. Since there are complex eigen-
values, from (4.28), the general solution is

x = d1 b 1 e t + d2 b 2 e t ,

where ! !
1 0
b1 = cos t − sin t,
−1 1
and ! !
1 0
b2 = sin t + cos t.
−1 1
Step 3: Satisfy the initial condition. Setting t = 0 in the general
solution, we get that
! ! !
1 0 0
d1 + d2 = .
−1 1 1
4.5. Summary for Solving a Homogeneous Equation 97

This can be written in component form as

d1 = 0
−d1 + d2 = 1.

So, d1 = 0 and d2 = 1.
Step 4: The resulting solution is
" ! ! #
1 0
x(t) = sin t + cos t et . 
−1 1

Example 3 (defective matrix): Find the solution of the IVP:

! !
′ 1 1 −1
x = x, where x(0) = .
−1 3 2

Step 1: Find the eigenvalues and eigenvectors. Using the eigenvalue

algorithm, from (4.14), you find the single eigenvalue r = 2, with
eigenvector !
1
a= .
1
To find a second independent solution, from (4.30) we must solve
! !
−1 1 1
b= .
−1 1 1

A solution of this is !
0
b= .
1
Step 2: Find the general solution. Using (4.29), the general solu-
tion is ! " ! !#
1 2t 1 0
x = c1 e + c2 t + e2t .
1 1 1
Step 3: Satisfy the initial condition. Setting t = 0 in the general
solution, we get that
! ! !
1 0 −1
c1 + c2 = .
1 1 2

This gives us c1 = −1 and c1 + c2 = 2. So, c2 = 3.

Step 4: The resulting solution is
" ! !#
1 −1
x = 3t + e2t . 
1 2
98 Chapter 4. Linear Systems

Exercises
1. Find a general solution of the following differential equations.
! !
−1 6 0 −9
a) x′ = x f) x′ = x
1 0 1 0
! !
1
0 4 1 5
b) x′ = x g) x′ = x
1 0 − 41 −1
! !
1
2 1 1 4
c) x′ = x h) x′ = x
6 3 −5 0
!
2 0
!
′
d) x = x 1 3
−1 2 i) x′ = x
−1 3
! !
−2 0 0 0
e) x′ = x j) x′ = x
0 −2 0 0

2. Find the solution of the initial value problem x′ = Ax, where the
differential equationis given in the previous problem, and the initial
4

condition is x(0) = −1 .
3. A solution of x′ = Ax is given below. What are the eigenvalues of A,
and what are corresponding eigenvectors?
! ! !
1 3t 1 e −2t
a) x(t) = e +2 et c) x(t) =
1 −1 3e4t
! ! !
1 −5t 1 e−8t − e−t
b) x(t) = e + d) x(t) =
0 3 3e−t

4. The general solution (4.25), and the eigenvalue algorithm given in Sec-
tion 4.3, can be used for any dimension n. In this exercise you are to
find the general solution for the case of when n = 3.
   
0 1 1 −2 2 0
a) x′ = 1 0 1x c) x′ =  0 1 0 x
   

1 1 0 4 2 −1
   
1 1 2 −1 0 0
b) x′ = 0 2 0x d) x′ =  0 1 2 x
   

0 1 1 0 2 −1
4.6. Phase Plane 99

4.6 Phase Plane

For differential equations involving 2×2 matrices, there are different ways
the solution can be portrayed. As an example, the general solution of the
differential equation !
2 1
x′ = x,
1 2
is ! !
1 3t 1
x(t) = c1 e + c2 et , (4.31)
1 −1
or, in component form,
x(t) = c1 e3t + c2 et ,
y(t) = c1 e3t − c2 et .
Given values for c1 and c2 , using the component form, graphing the solu-
tion simply involves plotting x and y as functions of t. In contrast, with
the vector version (4.31), the solution traces out a curve in the x,y-plane,
with t being the parameter that generates the curve. The x,y-plane is
referred to as the phase plane, and the curves that can be generated
using (4.31) are known as integral curves.

4.6.1 Examples
Two Positive Eigenvalues.

The solution (4.31) involves two positive eigenvalues, r1 = 3 and r2 =

1. The resulting integral curves generated by (4.31) are shown in Table
4.1(a). Each curve corresponds to a specific choice for c1 and c2 , and the
arrows indicate the direction for increasing t. Together, the integral curves
provide what is called a phase portrait for the equation. Any equation
with two positive eigenvalues will produce a phase portrait that is roughly
similar to the one for this example. A non-defective matrix with only one
eigenvalue, which is positive, will also have a roughly similar phase plane,
except the blue curves will be straight lines.
To explain how the phase portrait is constructed, you start by consid-
ering the c2 = 0 and the c1 = 0 cases first.
1
 
c2 = 0 : Since x = c1 1 e3t , then x = c1 e3t and y = c1 e3t . In other words,
y = x. This is the red line in Table 4.1(a) with positive slope. Be-
cause e3t increases with t, the solution moves outward, away from the
origin. So, the arrows on the line point outward. Note that the line
1

is determined by the eigenvector a1 = 1 , and the direction on the
line is determined by the positivity of the corresponding eigenvalue
r1 = 3.
100 Chapter 4. Linear Systems

a) r1 > 0, r2 > 0 b) r1 > 0, r2 < 0

source saddle
5 5
y-axis

y-axis
0 0

-5 -5
-10 -5 0 5 10 -10 -5 0 5 10
x-axis x-axis

c) r1 < 0, r2 < 0 d) r = ±iµ

sink center
5 5
y-axis

y-axis

0 0

-5 -5
-10 -5 0 5 10 -8 -4 0 4 8
x-axis x-axis

e) r = λ ± iµ with λ > 0 f) r = λ ± iµ with λ < 0

spiral source spiral sink
1 1
y-axis

y-axis

0 0

-1 -1
-1 0 1 -1 0 1
x-axis x-axis

Table 4.1. Examples of integral curves and how they depend on the eigenval-
ues of A. Each curve corresponds to a specific choice for the constants appearing in
the general solution. The arrows indicate the direction for increasing t. It is assumed
here that µ 6= 0.
4.6. Phase Plane 101

1
 
c1 = 0 : Since x = c2 −1 et , then x = c2 et and y = −c2 et . In other
words, y = −x. This is the red line in Table 4.1(a) with negative
slope. Because et increases with t, the solution moves outward, away
from the origin. So, the arrows on the line point outward. Note that
1

the line is determined by the eigenvector a2 = −1 , and the direc-
tion on the line is determined by the positivity of the corresponding
eigenvalue r2 = 1.

c1 6= 0 and c2 6= 0 : The general solution (4.31) consists of the addition of

the two components we just considered, and some of the resulting
integral curves are shown with the blue curves in Table 4.1(a). The
arrows on the curves point outward, away from the origin, because
both eigenvalues are positive. Also, since r1 > r2 , each solution curve
increases faster in the direction determined by a1 , and this is the
reason that the blue curves bend the way they do. Finally, note that
if you run time backwards, so t → −∞, then, from (4.31), x → 0.
That is why all of the blue curves look like they are emanating from
the origin. 

One Positive and One Negative Eigenvalue.

An example of this arises with the differential equation

!
′ −1 3
x = x,
2 0

which has eigenvalues r1 = 2 and r2 = −3. The general solution is found

to be ! !
1 2t 3
x(t) = c1 e + c2 e−3t . (4.32)
1 −2
The resulting integral curves are shown in Table 4.1(b). Any equation
with one positive, and one negative, eigenvalue will produce a phase por-
trait that is roughly similar to the one for this example.
As with the previous example, the phase portrait is constructed by
considering the c2 = 0 and the c1 = 0 cases first.
1
 
c2 = 0 : Since x = c1 1 e2t , then x = c1 e2t and y = c1 e2t . In other
words, y = x. This is the red line in Table 4.1(b) with positive
slope. Because e2t increases with t, the solution moves outward,
away from the origin. So, the arrows on the line point outward.
1

Note that the line is determined by the eigenvector a1 = 1 , and
the outward direction on the line is determined by the positivity of
the corresponding eigenvalue r1 = 2.
102 Chapter 4. Linear Systems

3
 
c1 = 0 : Since x = c2 −2 e−3t , then x = 3c2 et and y = −2c2 et . In other
words, y = −2x/3. This is the red line in Table 4.1(b) with negative
slope. Because e−3t decreases with t, the solution moves inward,
toward from the origin. So, the arrows on the line point  inward.
3

Note that the line is determined by the eigenvector a2 = −2 , and
the inward direction on the line is determined by the negativity of
the corresponding eigenvalue r2 = −3.

c1 6= 0 and c2 6= 0 : The general solution (4.31) consists of the addition of

the two components we just considered, and some of the resulting
integral curves are shown with the blue curves in Table 4.1(b). To
explain the arrows, the contribution of c2 a2 e−3t goes to zero as t
increases, but c1 a1 e2t becomes unbounded. A consequence is that a
solution curve will asymptotically approach the red line y = x. 

Two Negative Eigenvalues.

An example of this arises with the differential equation

!
′ −2 2
x = 1 x,
2 −2

which has eigenvalues r1 = −1 and r2 = −3. The general solution is

found to be ! !
2 −t −2 −3t
x(t) = c1 e + c2 e . (4.33)
1 1
The resulting phase portrait is shown in Table 4.1(c). Any equation with
two negative eigenvalues will produce a phase portrait that is roughly
similar to the one for this example. A non-defective matrix with one
eigenvalue, which is negative, will also have a roughly similar phase plane,
except the blue curves will be straight lines.
The construction of the phase portrait is very similar to what was done
for the two positive eigenvalues case. The principal difference is that the
eigenvalues are now negative, so the movement along the integral curves
is towards the origin. 

Imaginary Eigenvalues.

When the eigenvalues are imaginary, the integral curves are concentric
ellipses centered at the origin (see Exercise 5). To demonstrate this,
consider the differential equation
!
′ −2 4
x = x.
−2 2
4.6. Phase Plane 103

The eigenvalues are r1 = 2i and r2 = −2i, and the general solution, from
(4.28), is
" ! ! # " ! ! #
2 0 2 0
x(t) = d1 cos 2t− sin 2t +d2 sin 2t+ cos 2t . (4.34)
1 1 1 1

The ellipses generated by this solution are shown in Table 4.1(d).

The question is, is the movement around each ellipse clockwise, or
counter-clockwise? This can be determined from the differential equation.
For this example, x′ = −2x+4y, which means that when the ellipse crosses
the y-axis (so x = 0), x′ = 4y. Consequently, along the positive y-axis,
x′ > 0. The direction of the arrows must be consistent with this, and so
the rotation is clockwise. 

Complex Eigenvalues.

When the eigenvalues have nonzero real and imaginary parts the integral
curves are spirals centered at the origin (see Exercise 5). As an example,
!
′ 2 1
x = x,
−10 0

has eigenvalues r1 = 1 + 3i and r2 = 1 − 3i. The general solution, from

(4.28), is
" ! ! # " ! ! #
1 0 t 1 0
x(t) = d1 cos 3t − sin 3t e + d2 sin 3t + cos 3t et .
−1 3 −1 3
(4.35)
Similarly, for the differential equation
!
′ −2 1
x = x,
−10 0

the eigenvalues are r1 = −1 + 3i and r2 = −1 − 3i. The general solution

is
" ! ! # " ! ! #
1 0 1 0
x(t) = d1 cos 3t − sin 3t e−t + d2 sin 3t + cos 3t e−t .
1 3 1 3
(4.36)
The resulting integral curves for these two examples are shown in Table
4.1 (lower row). The one on the left comes from (4.35). The outward
motion in this case is because the real part of the eigenvalue is positive.
The one of the right comes from (4.36), and the inward motion is because
the real part of the eigenvalue is negative.
The spiral curves seen in these two graphs are explainable from the
formula for the solution. The solution contains cos µt and sin µt terms,
104 Chapter 4. Linear Systems

and these are responsible for the motion around the origin. This is similar
to what happens when r = ±iµ. However, these terms are multiplied by
eλt , and this causes the radial distance from the origin to either increase,
when λ > 0, or decrease, when λ < 0. 

4.6.2 Connection with an IVP

To illustrate the role the phase plane can play when solving an initial
value problem, suppose the problem to solve is
!
′ −1 3
x = x, (4.37)
2 0

where !
6
x(0) = . (4.38)
−3

This is the same differential equation used for the phase plane example in
Table 4.1(b), and the general solution is given in (4.32). From the initial
condition, the solution is found to be
! !
3 1 2t 9 3
x(t) = e + e−3t . (4.39)
5 1 5 −2

The plot of this curve in the phase plane is shown in Figure 4.1. The
integral curves for the differential equation, which appear in Table 4.1, are
also included in the figure. As this shows, the solution of the initial value
problem is simply a portion of one of its integral curves. The starting

5
y-axis

-5
-10 -5 0 5 10
x-axis

Figure 4.1. The solid blue curve is the solution (4.39), and the solid blue dot
is the location of the initial condition (4.38). The dashed blue curves, and the red lines,
are integral curves for (4.37).
Exercises 105

point is determined by the initial condition, and the resulting solution

follows the respective integral curve for increasing t.
The above observation is true in general. Namely, the integral curves
in Table 4.1 are illustrations of the various solutions you can get with the
respective differential equation. Which curve, or how much of the curve,
you get depends on the location of the initial condition.

Exercises
1. Phase portraits are shown in Figure 4.2, with arrows on some of the
curves. Do, or answer, the following: (i) Draw arrows on the other
curves. (ii) What properties of the eigenvalues result in the integral
curves shown in the phase portrait? (iii) Three different initial condi-
tions are shown by the black dots. For each one, sketch the solution
for the resulting IVP.
2. The eigenvalues for the following equations are real-valued. You are
to sketch the phase portrait as follows: (i) Draw the (red) lines that
are determined from the eigenvectors, and include the four arrows. (ii)
In each of the four quadrants determined by the red lines, include two
integral curves, with arrows.
! !
3 −1 3 2
a) x′ = x c) x′ = x
−1 3 −4 −3
! !
′
−6 3 ′
4 −2
b) x = x d) x = x
−4 1 3 −3

3. The eigenvalues for the following equations are imaginary. You are
to sketch the phase portrait as follows: draw three concentric ellipses
centered at the origin with arrows indicating the direction of motion.
It is useful to know that for A given in (4.4), when the eigenvalues
are imaginary, the elliptical integral curves are tilted right, as in Table
4.1(d), if ab < 0, and they are tilted left, as in Figure 4.2(e), if ab > 0.
! !
−1 2 2 1
a) x′ = x c) x′ = x
−2 1 −6 −2
! !
′
3 6 ′
−3 −1
b) x = x d) x = x
−3 −3 12 3

4. Spirals are either left (sinistral) or right (dextral) handed as shown in

Figure 4.3. For the matrix A given in (4.4), assume the eigenvalues
are λ ± iµ, with µ 6= 0.
106 Chapter 4. Linear Systems

5 5
y-axis

y-axis
0 0

-5 -5
-10 -5 0 5 10 -10 -5 0 5 10
x-axis x-axis
5 5
y-axis

y-axis
0 0

-5 -5
-10 -5 0 5 10 -10 -5 0 5 10
x-axis x-axis
5 5
y-axis

y-axis

0 0

-5 -5
-10 -5 0 5 10 -10 -5 0 5 10
x-axis x-axis

Figure 4.2. Integral curves, and location of three initial conditions, for Exercise 1.

a) Suppose that λ > 0. Using the same approach used to determine

clockwise or counter-clockwise motion for imaginary eigenvalues,
explain why a spiral integral curve is left-handed if b < 0 and it is
right-handed if b > 0. Also, explain why this reverses if λ < 0.
b) Determine if b for Figure 4.2(c) is positive or negative. What about
for Figure 4.2(d)?
5. This exercise involves the derivation of the formulas for the elliptical
and spiral curves obtained when the eigenvalues are complex-valued.
The matrix A is given in (4.4), and it is assumed that µ 6= 0.
a) Assuming that d1 and d2 are not both zero, from (4.28) and the
identity cos2 (µt) + sin2 (µt) = 1, show that

p22 x2 − 2p1 p2 xy + (p21 + q12 )y 2 = k 2 e2λt ,

4.7. Stability 107

y-axis

y-axis
0

0 0
x-axis x-axis

Figure 4.3. Left-handed (on the left) and right-handed (on the right) spirals.

where p1 = b(λ − a), p2 = (λ − a)2 + µ2 , q1 = −bµ, and k is a

positive constant.
b) In the equation in part (a), if λ = 0 you get an ellipse centered at
the origin. Explain why the movement along the ellipse is clockwise
if b > 0 and it is counter-clockwise if b < 0.
c) Show that to have a circular curve requires b 6= 0 and c = −b.
d) Explain why you get a spiral in the case of when λ 6= 0.

4.7 Stability
The phase plane can be useful for visualizing stability or instability of
a steady state solution. To explain how, recall from Section 2.4 that a
steady state is a constant that satisfies the differential equation. So, for
the equation x′ = Ax, a steady state is a constant vector xs that satisfies
Axs = 0. To avoid complications, it will be assumed that A is invertible,
which means that the only steady state solution is xs = 0. It is useful to
know that A is invertible if, and only if, r = 0 is not an eigenvalue for A.
The definitions of unstable and asymptotically stable are effectively
the same as in Section 2.4. Namely, a steady state xs is asymptotically
stable if any initial value x(0) chosen near xs results in
lim x(t) = xs . (4.40)
t→∞

The steady state is unstable if, no matter how close to xs you restrict
the choice for x(0), it is always possible to find an initial value x(0) that
results in the solution x(t) becoming unbounded as t increases.
It is easy to determine stability using the phase plane. For example,
in Table 4.1(a), when r1 > 0 and r2 > 0, the arrows on the integral
curves indicate movement out away from the origin. Consequently, this
is an example of when xs = 0 is unstable. Conversely, when r1 < 0 and
r2 < 0, the flow in towards the origin, and this means xs = 0 is asymp-
totically stable. In fact, looking at the various possibilities in Table 4.1,
you conclude that if A has an eigenvalue with Re(r) > 0, then the steady
state in unstable. Similarly, if the eigenvalues of A are both negative, or
if Re(r) < 0, then the steady state is asymptotically stable.
108 Chapter 4. Linear Systems

The conclusions in the previous paragraph were made using the phase
portraits in Table 4.1. For those that prefer more rigorous derivations,
then the formulas for the general solutions given in Section 4.5 can be
used.
Our classification of a steady state being unstable or asymptotically
stable does not include what happens when the eigenvalues are imaginary.
As shown in Table 4.1(d), the solution does not decay to zero, or blowup,
but simply encircles the origin. In this case, the steady state is said to be
neutrally stable.
The other case we are missing here is what happens when the matrix is
defective. From (4.29), the conclusion we had earlier still holds. Namely,
if r < 0, then we have asymptotically stability, and if r > 0, then we have
instability.
The above discussion is summarized in the following theorem.
Stability Theorem for a Linear System. For x′ = Ax, if r = 0 is
not an eigenvalue for A, then the following hold:
1. If all of the eigenvalues of A satisfy Re(r) < 0, then xs = 0 is an
asymptotically stable steady state.
2. If A has one or more eigenvalues with Re(r) > 0, then xs = 0 is
an unstable steady state.
3. If the eigenvalues of A are imaginary, then xs = 0 is a neutrally
stable steady state.
It is worth pointing out that the first two conclusions in the above theorem
hold when A is n × n. The third one also holds for the n × n case if
you make the additional assumption that A is not defective. For those
who might be wondering what happens when r = 0 is an eigenvalue,
the solution of Ax = 0 is no longer just xs = 0. In fact, any and all
eigenvectors for r = 0 are steady state solutions. It is possible to examine
the various cases that arise in this situation related to stability, but this
will not be considered in this text.
In addition to their stability, steady states are often identified by the
geometric properties of the solution near the steady state. So, for example,
because of the outward direction of the flow in Figure 4.1(a), the steady
state is called a source. In contrast, because of the inward flow in Figure
4.1(c), the steady state is called a sink. For similar reasons, the flow in
Figure 4.1(e) is a spiral source, and the one in Figure 4.1(f) is a spiral
sink. Finally, the steady state in Figure 4.1(b) is a saddle, and the one
in Figure 4.1(d) is a center.

Example 1: Determine the stability of the steady state xs = 0 for

!
′ 1 3
x = x.
2 −4
4.7. Stability 109

Answer: The characteristic equation for the matrix is r2 +3r −10 =

0, and from this it follows that the eigenvalues are r = −5 and r = 2.
Given that there is at least one eigenvalue that is positive, xs = 0
is unstable. Moreover, since it has one positive, and one negative,
eigenvalue, the steady state is a saddle point. 

Example 2: Determine the stability of the steady state xs = 0 for

!
−1 −2
x′ = x.
2 0

Answer: The characteristic equation for the matrix is r2 +r +4√= 0,

and from this it follows that the eigenvalues are r = 12 (−1 ± i 15).
Given that both have negative real part, then xs = 0 is asymp-
totically stable. Moreover, since the eigenvalues are complex with
negative real part, the steady state is a spiral sink. 

Example 3: Find the steady state, and determine its stability for
! !
′ 1 1 2
u = u+ . (4.41)
1 −1 4

Steady State: Since a steady state is a constant vector that satisfies

the differential equation, then we require that
! !
1 1 2
u=− .
1 −1 4

Solving this for u, one finds the steady state

!
−3
us = .
1

Stability: Letting u = us + x, and substituting this into the differ-

ential equation, one finds that x′ = Ax, where A is the matrix in
(4.41). If xs = 0 is unstable, then so is us . Similarly, if xs = 0
is asymptotically stable, then us asymptotically stable. Now, the
characteristic equation for A √is r2 − 2 = 0. From this, the eigen-
values are found to be r = ± 2. Given that one is positive, xs is
unstable, and therefore us is unstable. Moreover, since it has one
positive, and one negative, eigenvalue, us is a saddle point. 
110 Chapter 4. Linear Systems

Exercises
1. Determine whether xs = 0 is an asymptotically stable, unstable, or
neutrally stable steady state for the following differential equations.
Also, state whether the steady state is a sink, source, spiral sink, spiral
source, saddle, or center.
! ! !
−1 6 2 1 2 5
a) x′ = x d) x′ = x g) x′ = x
1 0 3 4 −5 −6
! ! !
1 2 −1 −1 0 −9
b) x′ = x e) x′ = x h) x′ = x
−3 −4 6 −6 1 0
! ! !
′
3 1 ′
1 14 ′
1 −4
c) x = x f) x = x i) x = x
1 3 −5 0 1 −1

2. Write the following as x′ = Ax, and then determine whether xs = 0

is an asymptotically stable, unstable, or neutrally stable steady state.
a) The simple harmonic oscillator given in (3.49).
b) The damped oscillator given in (3.57).
3. Find the steady state us , and determine its stability, for the following
differential equations. Also, state whether the steady state is a sink,
source, spiral sink, spiral source, saddle, or center.
! ! ! !
1 3 1 −3 −1 1
a) u′ = u+ c) u′ = u−
0 −1 0 2 −1 2
! ! ! !
−2 1 −2 1 −1 1
b) u′ = u+ d) u′ = u+
1 −2 1 4 1 −1

4. This exercise contains useful information to determine the stability of

xs = 0 without having to calculate eigenvalues. Assume that A is
given in (4.4) and that det(A) 6= 0. Also, the trace of a matrix is the
sum of the numbers on the diagonal. The formula is tr(A) = a + d.
a) Show that the eigenvalues of A are 12 tr(A)± [tr(A)]2 − 4det(A) .
 p 

b) Explain why r = 0 is not an eigenvalue for A.

c) Show that if tr(A) > 0, then xs is unstable.
d) Show that if det(A) < 0, then xs is unstable.
e) Show that if tr(A) = 0, and det(A) > 0, then xs is neutrally stable.
f) Show that if tr(A) < 0 and det(A) > 0, then xs is asymptotically
stable.
Chapter 5

Nonlinear Systems

This chapter considers problems that involve two first-order ordinary

differential equations, at least one of which is nonlinear. These problems
are usually difficult enough that finding a formula for the solution is not
possible. Consequently, most of the chapter does not concern solving these
problems, but instead concentrates on developing ways to determine the
properties of the solution. What this means exactly will be explained
as the methods are derived. We begin with examples that illustrate the
problems we will be considering.

Example 1: Pendulum
The equation for the angular deflection of a pendulum is (see Figure 5.1)

d2 θ
ℓ = −g sin θ, (5.1)
dt2

where the initial angle θ(0) and the initial angular velocity θ′ (0) are as-
sumed to be given. Also, ℓ is the length of the pendulum and g is the grav-
itational acceleration constant. Introducing the angular velocity v = θ′
then the equation can be written as the first-order system

θ′ = v, (5.2)
′
v = −α sin θ, (5.3)

where α = g/ℓ. Although (5.2) is linear, (5.3) is nonlinear because of the

sin θ term. Consequently, together (5.2), (5.3) form a nonlinear first-order
system for θ and v. 

Introduction to Differential Equations, M. H. Holmes, 2020

111
112 Chapter 5. Nonlinear Systems

Figure 5.1. Angular deflection of a pendulum.

Example 2: Measles
A model for the spread of a disease, like measles, is

dS
= αN − (βI + α)S,
dt
dI
= βIS − (α + γ)I.
dt

In these equations, S(t) is the number of people susceptible to the disease,

and I(t) is number that are ill. The nonlinearity, which is due to the term
IS, appears in both equations. 

The equations for the pendulum and the spread of measles are not
solvable using elementary functions. What is possible it to ask questions
about the solution that are significant and answerable. As an example,
with measles, a reasonable question would be: what would it take to
eliminate the disease from the population? This requires that I → 0 as
t → ∞ (and the faster this happens the better). In more mathematical
terms, we want I = 0 to be an asymptotically stable steady state. How
to modify the stability of I = 0, with the goal of quickly eliminating the
disease, will be considered in Section 5.2.2.
A question arising with the pendulum is, does it ever stop moving?
Given the physical assumptions used in the derivation of the equation it
is reasonable to expect that it does not stop and, in fact, the solution
is expected to be periodic. So, we would like to know if it is possible
to show that the solution is periodic, and in the process determine the
period (without actually solving the problem).

5.1 Non-Linear Systems

The problems in this chapter can be written in component form as

u′ = f (u, v), (5.4)

′
v = g(u, v). (5.5)
5.1. Non-Linear Systems 113

In these equations, u(t) and v(t) are the dependent variables, and f and
g are given functions of u and v. It is assumed that the equations are
autonomous, which means that f and g do not depend explicitly on t.
The vector form of (5.4), (5.5) is
dy
= f (y), (5.6)
dt
where ! !
u f (u, v)
y= , and f= .
v g(u, v)
For an initial value problem, an initial condition of form
!
u0
y(0) = . (5.7)
v0
would also be given.

Example: For the nonlinear system

1
u′ = v − u, (5.8)
2
1
v ′ = − v + 2u(2 − u2 ), (5.9)
2
we have that !
v − 21 u
f= .
− 12 v + 2u(2 − u2 )
There are no known mathematical methods that can be used to find
the solution of this system (by hand). However, it is easily solved
using a computer, and four example curves are shown in Figure 5.2.
In all four cases, the solution ends up at one of two points. In this
chapter we will not attempt to find the solution curves but we will
be very interested in determining these two points and finding the
reason why the solution approaches them. 

5
v-axis

-5
-3 -2 -1 0 1 2 3
u-axis

Figure 5.2. Solution curves of (5.8), (5.8) in the u,v-plane for four different
initial conditions (shown with the solid dots). The arrows indicate the direction for
increasing t.
114 Chapter 5. Nonlinear Systems

5.1.1 Steady-State Solutions

For y′ = f (y), a steady state solution ys is a constant vector that
satisfies f (ys ) = 0. In component form, the requirements are that

f (us , vs ) = 0, (5.10)
g(us , vs ) = 0. (5.11)

Solving for us and vs is not straightforward. In fact, given that f (u, v)

and g(u, v) can be almost anything, there is no method that always works
for solving these equations. The recommendation is to pick one of the
equations, and use it to solve for u in terms of v, or v in terms of u. The
equation to pick for this is usually the one that is easiest to solve. This
solution is then substituted into the other equation, and you then have
one equation and one unknown (see Example 1). It is also not uncommon
that you need to be opportunistic, and take advantage of certain terms
in the equation to help simply the equations (see Example 2).

Example 1: Find the steady states of

du
= 3 − u − v − uv,
dt
dv
= uv − 2v.
dt
Answer: The equations to solve are

3 − u − v − uv = 0,
uv − 2v = 0.

The second equation looks the easiest to work with. Factoring it as

v(u − 2) = 0, we get two solutions: v = 0 and u = 2. Taking v = 0,
then from the first equation we get that u = 3. For u = 2, from the
first equation we get that v = 1/3. Therefore, we have found two
steady states: (us , vs ) = (3, 0), and (us , vs ) = (2, 1/3). 

Example 2: Assuming α is a positive constant, find the steady states of

du
= 1 − (1 + α)u + u2 v,
dt
dv
= u − u2 v.
dt
Answer: The equations to solve are

1 − (1 + α)u + u2 v = 0,
u − u2 v = 0.
Exercises 115

It is possible to use the approach from the previous example, but it

is easier to look a little closer at these equations. They both contain
the term u2 v. In fact, from the second equation u2 v = u. Using this
information in the first equation, we get that u = 1/α. From the
second equation, it follows that v = α. Therefore, we have found
that the only steady state is: (us , vs ) = (1/α, α). 

Example 3: Find the steady states of

x′ = x − x2 − xy,
y ′ = 2y − y 2 − 3xy.

Answer: The equations to solve are

x − x2 − xy = 0,
2y − y 2 − 3xy = 0.

Factoring the first equation as x(1 − x − y) = 0, then either x = 0 or

x = 1 − y. If x = 0, then from the second equation y = 0 or y = 2,
giving us the two steady states (0, 0) and (0, 2). When x = 1 − y,
the second equation reduces to y(1 − 2y) = 0, which has solutions
y = 0 and y = 1/2. This gives us two more steady states, which are
(1/2, 1/2) and (1, 0). 

Example 4: For the system

x′ = x − y,
y ′ = (x − y)3 ,

the steady states are any points that satisfy y = x. 

We are going to avoid the situation in Example 4. Specifically, in

the problems we will consider, there can be multiple steady states, but
they are discrete points as in Examples 1, 2, and 3. The way this will be
stated is that the steady states are isolated, which means that there is
a nonzero distance d so that the distance between any two steady states
for the problem is at least d.

Exercises
1. Write the following as y′ = f (y), making sure to identify the entries in
y and f . If initial conditions are given, write them as y(0) = y0 .
116 Chapter 5. Nonlinear Systems

a) u′ = u2 − v g) Michaelis-Menten system
v ′ = 2u − 3v S ′ = −k1 ES + k−1 (E0 − E),
b) u′ = u2 + v 2 E ′ = −k1 ES+(k2 +k−1 )(E0 −E)
2v ′ = sin(u) S(0) = 1, E(0) = 2
c) u′ = eu − v h) Predator-prey
v ′ = uv x′ = ax − bxy
u(0) = −1, v(0) = 0 y ′ = −cy + dxy
d) Van der Pol oscillator i) Projectile (nonuniform field)
gR2
u′′ + (1 − u2 )u′ + u = 0 y ′′ = −
(R + y)2
e) Toda oscillator y(0) = 0, y ′ (0) = 3
u′′ + eu − 1 = 0
j) Orbital motion
f) Duffing oscillator α2 µ
u′′ + u + u3 = 0 r′′ = 3 − 2
r r
u(0) = 1, u′ (0) = −1 r(0) = 1, r′ (0) = 2

2. Find the steady state solutions of the following.

( (
u′ = 1 − 2u − v − uv S ′ = −IS + 5 − I − S
a) e)
v ′ = 3uv − v I ′ = IS − I
( (
u′ = v − u2 s′ = c − s2
b) f)
v ′ = v + u3 c′ = 1 + sc
( (
u′ = 4 − uv 2 x′ = sin(y) + sin(x)
c) g)
v ′ = −v + uv 2 y ′ = 3y 2 + x4
( (
2SP
S ′ = 2S − S 2 − 1+S x′ = xy
d) 2SP h)
P ′ = 1+S −P y ′ = (2 − x − y)(1 + y)

5.2 Stability
The question considered now is central to this chapter, and it is whether
a steady state is achievable. What this means is that the steady state is
asymptotically stable. To explain how we are going to determine stability,
consider the problem of solving

x′ = x − x2 − xy, (5.12)
′ 2
y = 2y − y − 3xy. (5.13)

This is the problem from Example 3 in the previous section, and we found
that there are four steady states: (0, 0), (0, 2), (1, 0), and (1/2, 1/2). One
approach for providing insight about stability is to solve the problem
numerically. This is easy to do, and two computed solution curves are
shown in Figure 5.3. The curves are consistent with what is expected if
5.2. Stability 117

2
y-axis
0

-2

-4

-3 -2 -1 0 1 2 3 4
x-axis

Figure 5.3. Solution of (5.12),(5.13) for different initial conditions. The

blue curve approaches the steady state (1, 0), while the red curve approaches the steady
state (0, 2).

(0, 2) and (1, 0) are asymptotically stable. Also, since both curves start
near (0, 0), yet move away from it, it would not be a surprise to find out
that (0, 0) is an unstable steady state.
Solving the problem numerically is so easy that it possible to solve the
problem for many different initial conditions, and check if the solution
approaches one of the various steady states. The results from such a
calculation are shown in Figure 5.4. What is found is that there are,
apparently, two asymptotically stable steady states, (0, 2) and (1, 0). The
calculations also identity the regions for the initial conditions that result
in the solution ending up at the respective steady state. The two regions
determined from this computation are called the domain of attraction for
the respective steady state.
Our goal is not to be able to determine the shaded regions shown
in Figure 5.4, but, rather, to show that there is a small region around
the respective steady state with the same property as the shaded region.
Namely, for any initial condition in that small region, the solution of the
resulting IVP will end up at the steady state. In this case, the steady
state is said to be asymptotically stable. What we are doing now is the
two dimensional version of what we did in Section 2.4, and the nonlinear
version of what was done in Section 4.7.

5.2.1 Derivation of the Stability Conditions

The differential equation is y′ = f (y), and this can be written in compo-
nent form as

u′ = f (u, v), (5.14)

′
v = g(u, v). (5.15)
118 Chapter 5. Nonlinear Systems

Figure 5.4. An initial condition (x(0), y(0)) located in one of the shaded
regions results in the solution of (5.12),(5.13) ending up at the steady state in that
shaded region. The two steady states are shown by the dark circles.

Assume that (us , vs ) is a steady-state, which means that us and vs are

constants that satisfy

f (us , vs ) = 0,
g(us , vs ) = 0.

The reason for considering stability comes from this question: If we start
the solution near (us , vs ), what happens?
There are three possible conclusions coming from this question: the
steady state is unstable, it is asymptotically stable, or it is neutrally
stable. What these are can be explained using a ball and bowl (see Figure
5.5). The force on the ball is gravity. For the bowl, the steady state is at
the bottom, and for the inverted bowl it is at the top. For the inverted
bowl, if you release the ball from rest, no matter where you place it (other
than exactly at the top), the ball will roll away. The conclusion is that
the steady state is unstable. For the bowl, you can control how far the
ball will get from the steady state (the bottom) by placing it close to
the bottom and giving it only a small initial velocity. Consequently, the
steady state is stable. Because the only force is gravity, the ball will roll
around in the bowl forever. This means the steady state is neutrally
stable. If the problem also includes damping, such as friction, then the
ball will slow down and eventually come to rest at the bottom. In this
case the steady state is asymptotically stable. Note that including

Figure 5.5. Ball in a bowl, on the left, and a ball on an inverted bowl, on the right.
5.2. Stability 119

damping for the inverted bowl will not change the fact that the top is an
unstable steady state.
For those that prefer a more mathematical definition, the idea under-
lying asymptotic stability is that if y(0) is any point close to the steady
state ys , then
lim y(t) = ys . (5.16)
t→∞

As stated above, a steady state is stable if you can control how far the
solution gets from ys by picking y(0) close to ys . Specifically, given any
ε > 0, you can find a δ > 0 so that if ||y(0)−ys || < δ, then ||y(t)−ys || < ε.
If this is not possible then ys is unstable. If ys is stable, and (5.16) holds,
then it is asymptotically stable. Otherwise it is said to be neutrally stable.
Note that this version of the definition of stability√requires that u and v
have the same physical dimensions so that ||y|| = u2 + v 2 is defined.
To answer the stability question, assume that the initial position
(u(0), v(0)) is very close to (us , vs ). To determine what happens, we will
use what is called the linear approximation in multivariable calculus. This
states that if f (u, v) and g(u, v) are differentiable at (us , vs ), then each
can approximated using their respective tangent plane. In particular,

f (u, v) ≈ f (us , vs ) + fu (us , vs )(u − us ) + fv (us , vs )(v − vs ),

g(u, v) ≈ g(us , vs ) + gu (us , vs )(u − us ) + gv (us , vs )(v − vs ).

In the above expressions, fu = ∂f ∂f ∂g ∂g

∂u , fv = ∂v , gu = ∂u , gv = ∂v . It should
be pointed out that this approximation is also a direct consequence of
Taylor’s theorem, and this can be used to derive more accurate approxi-
mations if needed.
By assumption, f (us , vs ) = 0 and g(us , vs ) = 0. Consequently, the
linear approximation of (5.14) and (5.15) near the steady state is

u′ = fu (us , vs )(u − us ) + fv (us , vs )(v − vs ),

v ′ = gu (us , vs )(u − us ) + gv (us , vs )(v − vs ).

This can be written in system form as

y′ = J(y − ys ), (5.17)

where ! !
u us
y= , ys = ,
v vs
and !
fu (us , vs ) fv (us , vs )
Js = .
gu (us , vs ) gv (us , vs )
The matrix Js is known as the Jacobian matrix of f evaluated at ys .
120 Chapter 5. Nonlinear Systems

To put the problem into the form covered in the last chapter, let
x = y − ys . With this, (5.17) becomes

x′ = Ax, (5.18)

where A = Js . The general solution of this is given in Section 4.5. For

what we are doing it is not necessary to distinguish between real or com-
plex valued eigenvalues. Using the formulas in Section 4.5, and remem-
bering that y = ys + x, we conclude that if Js is not defective, then

y = ys + c1 a1 er1 t + c2 a2 er2 t , (5.19)

and if it is defective, then

y = ys + c1 aert + c2 (ta + b)ert . (5.20)

Whether the ert terms in (5.19) or (5.20) go to zero, or blow up, as

t → ∞, depends on whether Re(r) is positive or negative. To determine
this, it is easiest to go through the various possibilities individually.

• If all of the eigenvalues of Js satisfy Re(r) < 0, then the exponentials

in (5.19) and (5.20) go to zero as t → ∞. So, ys is asymptotically
stable

• If one, or more, of the eigenvalues of Js satisfies Re(r) > 0, then

at least one of the exponentials in (5.19) and (5.20) blows up as
t → ∞. So, ys is unstable.

There is a notable hole in the above list in that there is no conclusion for
the case of when the eigenvalues are imaginary. In the theorem in Section
4.7, this is referred to as being neutrally stable. There are neutrally stable
steady states for nonlinear systems, as illustrated with the ball and bowl
example earlier, but the tangent plane approximation is inadequate to
determine this. One approach to show neutral stability is to use the ideas
developed in Section 5.3.
As a final comment, the only assumption needed to guarantee that the
above conclusions hold is that the first and second partial derivatives of
f (u, v) and g(u, v) are continuous. Those interested in a mathematically
rigorous proof of this should consult Stuart and Humphries [1998] or Perko
[2001].

Phase Plane

The above derivation for the stability conditions can provide us with in-
formation about the solution curves near a steady state. The reason is
that the reduced equation in (5.18) is the same one considered in the last
5.2. Stability 121

chapter. This enables us, in certain cases, to apply the phase plane solu-
tions shown in Table 4.1 (page 100) to the nonlinear system. To explain
how, suppose you have a steady state that the above test determines is
unstable or asymptotically stable. As stated earlier, we are only consid-
ering isolated steady states, and to guarantee this happens it is assumed
that r = 0 is not an eigenvalue. Now, in the vicinity of the steady state,
we have that y ≈ ys +x. This means that the phase portrait for y is simi-
lar to one of those in Table 4.1, but it is centered at y = ys rather than at
x = 0. Which one is determined by the eigenvalues of Js . Demonstrations
of this will be included in the examples that follow.

5.2.2 Summary
For the nonlinear system

u′ = f (u, v)
v ′ = g(u, v),

the associated Jacobian matrix J is given as

∂f ∂f
 
 ∂u ∂v 
J= ∂g
.
∂g 
∂u ∂v
The eigenvalues of J are used to determine stability, as explained in the
next theorem.

Linearized Stability Theorem. Given a steady state ys , and letting

Js be the Jacobian matrix evaluated at ys :
• If all of the eigenvalues of Js satisfy Re(r) < 0, then ys is asymp-
totically stable.
• If one, or more, of the eigenvalues of Js satisfies Re(r) > 0, then
ys is unstable.
This assumes that the second partial derivatives of f (u, v) and g(u, v) are
continuous at, and in the immediate vicinity of, ys .

Not every possibility is included in the above theorem. As an example, no

conclusion can be made when there are only imaginary eigenvalues. Any
case that is not covered by the theorem will be referred to as indeterminate
in this chapter.
For those with good memories, there are a few easy to use shortcuts
that avoid computing eigenvalues. If you are interested in what they are,
see Exercise 4.
122 Chapter 5. Nonlinear Systems

It is worth pointing out that even though we are considering systems

with two equations (so, n = 2), the above theorem holds when there are n
equations. In fact, for n = 1 the above theorem reduces to the one given
in Section 2.4.1 (page 35).
Finally, if the above theorem determines that a steady state is un-
stable or asymptotically stable, and r = 0 is not an eigenvalue, then the
eigenvalues and eigenvectors of Js can be used to determine the phase
portrait of the solution near the steady state. This is done in the same
way as for the examples shown in Table 4.1. The principal difference now
is that it is centered at y = ys rather than at x = 0. Therefore, the clas-
sification of steady states into a source, sink, spiral source, spiral sink,
or saddle, as given on page 108, is applicable to the nonlinear systems
considered here.

5.2.3 Examples

Example 1: Determine the stability of the steady states of

1
u′ = v − u,
2
1
v ′ = − v + 2u(2 − u2 ).
2
This is the system that produced the solution curves shown in Figure
5.2.
Step 1: Find the steady states. The equations to solve are
1
v − u = 0,
2
1
− v + 2u(2 − u2 ) = 0.
2
One finds that there are three steady √ states, and they are: (u, v) =
−(2α, α), (0, 0), (2α, α), where α = 18 30.
Step 2: Determine the Jacobian matrix.
∂f ∂f
 
!
 ∂u ∂v  − 12 1
J= = .
 ∂g ∂g  2(2 − 3u2 ) − 12
∂u ∂v

Step 3: Check each steady state.

(2α, α): In this case !

− 21 1
Js = ,
− 29
4 − 21
5.2. Stability 123

√ √
and this has eigenvalues r1 = (−1 + i 29)/2 and r2 = (−1 − i 29)/2.
Since both satisfy Re(r) < 0, this steady state is asymptotically stable.
In addition, since the eigenvalues are complex, and Re(r) < 0, the
phase portrait near this steady state should be a spiral sink. To check,
the region in Figure 5.2 that is near (2α, α) is shown in Figure 5.6.
As expected, the solution curves spiral into the steady state, as they
should for a spiral sink.
(0, 0): In this case !
− 21 1
Js = ,
4 − 12
and this has eigenvalues r1 = 3/2 and r2 = −5/2 . Since r1 > 0 then
this steady state is unstable. Also, since r2 < 0 < r1 , then this is a
saddle point and the phase portrait near (0, 0) will resemble the one in
Figure 4.1(b) or in Figure 4.2(b).
−(2α, α): In this case !
− 12 1
Js = ,
− 29
4 − 12
√ √
and this has eigenvalues r1 = (−1 + i 29)/2 and r2 = (−1 + i 29)/2.
Since both satisfy Re(r) < 0, this steady state is asymptotically stable.
As with (2α, α), this is a spiral sink. 

Example 2: Determine the stability of the steady states of

x′ = x − x2 − xy,
y ′ = 2y − y 2 − 3xy.

This is the system that produced the solution curves shown in Figure
5.3.
Answer: In Section 5.1.1, Example 3, we found that there are four

2
v-axis

-0.3
0.6 1 1.4
u-axis

Figure 5.6. Solution curves of (5.8), (5.8) in the u,v-plane near the steady
state (2α, α).
124 Chapter 5. Nonlinear Systems

steady states: (0, 0), (0, 2), (1/2, 1/2) and (1, 0). To determine their
stability, the Jacobian is
 ∂f ∂f  !
 ∂x ∂y  1 − 2x − y −x
J=  ∂g
= .
∂g  −3y 2 − 2y − 3x
∂x ∂y

(0, 2): In this case !

−1 0
Js = .
−6 −2
The eigenvalues are r1 = −1 and r2 = −2, and since they are both neg-
ative, the steady state is asymptotically stable. Moreover, since both
are negative, the phase portrait near this steady state will resemble
those for a sink. Sketching the phase portrait was explained in Section
4.6. Briefly, eigenvectors of Js , for r1 and r2 are, respectively,
! !
−1 0
a1 = , and a2 = .
6 1

The two red lines shown in Figure 5.7 are determined by these eigen-
vectors. The arrows point toward the steady state as both eigenvalues
are negative. Typical integral curves are shown in blue. The result is
a phase portrait for a sink.
(1/2, 1/2): In this case
!
−1/2 −1/2
Js = ,
−3/2 −1/2
√ √
and this has eigenvalues r1 = (−1+ 3)/2 and r2 = (−1− 3)/2. Since
r1 > 0, it follows that this steady state is unstable. As for the phase
portrait near this steady state, since r2 < 0 < r1 , then this steady

2.1
y-axis

1.9
-0.02 0 0.02
x-axis

Figure 5.7. Phase portrait near the steady state (0, 2) for Example 2.
5.2. Stability 125

0.52

y-axis 0.5

0.48
0.48 0.5 0.52
x-axis

Figure 5.8. Solution curves of Example 2 in the x,y-plane near the steady
state (1/2, 1/2).

state is a saddle point. To sketch the phase portrait, the eigenvectors

of Js , for r1 and r2 are, respectively,
√ ! √ !
− 31 3 1
3 3
a1 = , and a2 = .
1 1

The two red lines determined by these vectors are shown in Figure
5.8. Typical integral curves are shown in blue. So, the curves have the
pattern expected for a saddle.
Determining the stability of the remaining two steady states is left as
an exercise. 

Example 3: As introduced at the beginning of the chapter, a model for

the spread of a disease, like measles, is
dS
= αN − (βI + α)S,
dt
dI
= βIS − (α + γ)I,
dt
where N is the total number of individuals in the population (it is
constant). The coefficients, α, β, and γ, are positive constants. It is
not hard to show that the two steady states are (S, I) = (N, 0) and
(S, I) = (Se , Ie ), where
α+γ α
Se = and Ie = (N − Se ) .
β α+γ
The first steady state, (N, 0), corresponds to the case of when the dis-
ease is eliminated, and everyone ends up in the S group. The other
steady state, (Se , Ie ), is an example of what is known as an epidemic
equilibrium, and this is something that is usually avoided if at all pos-
sible. Said another way, we want this steady state to be unstable.
126 Chapter 5. Nonlinear Systems

To determine the stability of the steady states, note that

∂f ∂f
 
!
 ∂S ∂I  −(βI + α) −βS
J= = .
 ∂g ∂g  βI βS − (α + γ)
∂S ∂I

(S, I) = (N, 0): In this case

!
−α −βN
Js = .
0 βN − (α + γ)

The eigenvalues of this matrix are −α and β(N − Se ). Therefore, if

N < Se , then this steady state is asymptotically stable, and if N > Se ,
then it is unstable.

(S, I) = (Se , Ie ): One finds that this steady state is unstable if N < Se ,
and it is asymptotically stable if N > Se .

Measles: According to the model, to eradicate the disease, which means

that (Se , Ie ) is unstable, it is required that
α+γ
N< . (5.21)
β
The parameter α is the birth rate in the population and γ is associated
with the rate at which people get well, both of which you can do little
to change. As for β, it reflects how contagious the disease is (a larger
β means it is more contagious). For measles, α = 1/50, γ = 100, and
β = 1800/N [Engbert and Drepper, 1994], in which case
α+γ 1
≈ N.
β 18
Clearly, (5.21) is not even close to being satisfied. This is a reflection
of that fact that measles is one of the most contagious diseases known.
What is needed is to reduce β by a factor of 20 (or more). It is possible
to make β smaller by taking actions that limit the propagation of the
disease, but finding effective ways to do this is challenging. 

Exercises
1. For the following find the steady states, and then determine whether
they are asymptotically stable, unstable, or indeterminate. Also, ex-
cept for the indeterminate cases, state whether the steady state is a
sink, source, spiral sink, spiral source, or saddle. Any parameters ap-
pearing in the equations should be assumed to be positive.
Exercises 127

( (
u′ = 1 − 2u − v − uv x′ = ex − y
a) g)
v ′ = 3uv − v y ′ = xy
(
( x′ = ax − bxy
u′ = v − u h)
b) y ′ = −cy + dxy
v ′ = v + u3
(
2SP
(
u′ =1+v S ′ = 2S − S 2 − 1+S
c) i) 2SP
v′ = u + v3 P′ = 1+S −P
(
S ′ = − 21 IS + 1 − I − S
(
u′ =4− uv 2 j)
d) I ′ = 12 IS − I
v ′ = −v + uv 2
(
r′ = s − r
(
x′ = x2 − y
e) k)
y ′ = 2x − 3y s′ = (2 − r − s)(1 + s2 )
( (
x′ = x2 + y 2 S ′ = −2ES + E0 − E
f) l)
2y ′ = sin(x) E ′ = −2ES + 2(E0 − E)

2. For the following: (i) find the steady state, (ii) find the linear approx-
imation of the system near the steady state, and then (iii) sketch the
phase portrait in the vicinity of the steady state as follows: draw the
(red) lines that are determined from the eigenvectors of Js , including
the arrows for these lines, then in each of the four quadrants deter-
mined by the red lines, include two integrals curves, with arrows.
( (
u′ = v − u r′ = s − r
a) c)
v ′ = v + u3 s′ = (2 − r − s)(1 + s2 )
( (
u′ = 1 + v S ′ = −2ES + E0 − E
b) d)
v′ = u + v3 E ′ = −2ES + 2(E0 − E)

3. Suppose that y = Y is a steady state solution of y ′′ + cy ′ + g(y) = 0.

So, y = Y is a constant and g(Y ) = 0.
a) Show that Y is unstable if c < 0.
b) Show that Y is asymptotically stable if c > 0 and g ′ (Y ) > 0, and it
is unstable if c > 0 and g ′ (Y ) < 0.
4. In this problem, assume that
!
a b
Js = .
c d

The trace of a matrix is the sum of the numbers on the diagonal. The
formula is tr(Js ) = a + d. Also, the determinant is det(Js ) = ad − bc.
128 Chapter 5. Nonlinear Systems

h i
1
p
a) Show that the eigenvalues of Js are 2 tr(Js )± [tr(Js )]2 − 4det(Js ) .
b) Show that if tr(Js ) > 0, then ys is unstable.
c) Show that if det(Js ) < 0, then ys is unstable.
d) Show that if det(Js ) > 0 and tr(Js ) < 0, then ys is asymptotically
stable.
5. This exercise considers the curve, in the first quadrant, that separates
the red and blue regions in Figure 5.4.
a) Explain why the curve must contain the point (1/2, 1/2).
b) Suppose that the initial point (x(0), y(0)) is on the curve. Explain
why the resulting solution (x(t), y(t)) must remain on the curve.
6. A model for how a joke moves through a population involves three
groups: S is the population that either has not heard the joke, or does
not remember it, T is the population of those who know the joke and
they will tell it to others, and R is the population who know the joke
but will not tell it to others (they are not good joke tellers or they
don’t think it’s all that funny). As shown in Holmes [2019],

dS
= −2αST + β(N − S),
dt
dT
= αST − βT,
dt
where N is the total number of individuals in the population (it is
constant). The coefficients α and β are positive constants. Also, once
S and T are determined, then R = N − T − S.
a) There are two steady states, what are they?
b) One of the steady states has T = 0. When is it asymptotically
stable?
c) One of the steady states has T 6= 0. When is it asymptotically
stable?
d) The α is the telling parameter, so a larger α means the joke is
being told more often. Similarly, β is the forgetting parameter, so
a larger β means the joke is being forgotten faster. Based on your
answers from parts (b) and (c), under what conditions will the joke
disappear from the population?

5.3 Periodic Solutions

With the stability test derived in the previous section, we have a fairly
good tool for determining if, and when, the solution of a nonlinear system
will come to rest. The next question concerns what can be learned about
periodic solutions. This is needed as periodicity plays an important role
5.3. Periodic Solutions 129

in our lives, and examples are the sleep-wake cycle and the periodic events
associated with the Earth’s rotation.
To begin, it’s best to define what is meant by periodicity. A solution
of y′ = f (y) is periodic if there is a positive number T so that
y(t + T ) = y(t), ∀t ≥ 0. (5.22)
The smallest positive T , if it exists, is the period.
We will only consider problems that come from Newton’s second law.
Specifically, if u(t) is the displacement, and F is a function of u, then
F = ma gives us the differential equation
mu′′ = F (u). (5.23)
Letting v = u′ , then the above equation can be written in system form as
u′ = v, (5.24)
1
v ′ = F (u). (5.25)
m
It is not hard to show that if u(t) is periodic with period T , then the
velocity v(t) is also periodic with period T . Consequently, (5.22) is sat-
isfied, and so the solution is periodic. Examples of what are, or are not,
periodic are explored in more depth in Exercise 2.
We will first find a way to determine the solution curve in the phase
plane directly from the differential equation and initial conditions. Once
that is done, we will then be able to determine the period T , as well as
other properties of the solution.

Example: Mass-Spring
In Section 3.10, it was shown that the displacement u(t) of a mass in
a spring-mass system satisfies mu′′ + ku = 0. The general solution of
this equation can be written as p u = R cos(ω0 t − ϕ), and v = u′ =
−ω0 R sin(ω0 t − ϕ), where ω0 = k/m. Consequently, the solution is
periodic, with period T = 2π/ω0 . The key observation here is that, using
the identity cos2 θ + sin2 θ = 1,
 u 2  v 2
+ = 1,
R ω0 R
or equivalently
1 2
u2 + v = R2 . (5.26)
ω02
This is an equation for an ellipse in the u,v-plane. As an example, suppose
that m = 1, k = 4, and the initial conditions are u(0) = 1 and v(0) = 0.
In this case, u = cos(2t), v = −2 sin(2t), and from (5.26), the ellipse is
1
u2 + v 2 = 1. (5.27)
4
130 Chapter 5. Nonlinear Systems

v-axis
0

-1

-2
-2 -1 0 1 2
u-axis

Figure 5.9. Elliptical path, given in (5.27), that is followed by the solution
of the mass-spring IVP. The blue dot is the location of the initial condition.

This curve is shown in Figure 5.9. Because the period is T = π, the

solution goes around the ellipse and returns to the starting point (1, 0) at
t = π, 2π, 3π, . . ..
To see what can be learned from the system form of the problem, the
equations are

u′ = v,
v ′ = −ω02 u.

This can be used to determine the direction of the arrows in Figure 5.9.
Since v ′ = −ω02 u, using the initial condition given earlier, v ′ (0) = −ω02 .
The fact that this is negative means that v must decrease as it leaves
the initial point, and so the direction of motion is clockwise around the
curve. Note that it is not possible for the solution to reverse direction on
the curve because this would require that there is a point on the curve
where u′ = 0 and v ′ = 0. Such a point corresponds to a steady state, and
the only steady state for this problem is the origin. 

The important conclusion coming from the above example is that, no

matter what time t you select, the solution is located somewhere on the
curve shown in Figure 5.9. Having a closed curve like this is a requirement
for the solution to be periodic. The reason is that a solution traces out
a curve in the phase plane, whether the solution is periodic or not (see
Table 4.1 for examples). For the solution to be periodic, it must return
to its original position, and that is why a closed curve as in Figure 5.9 is
required. What is shown below is how to determine this curve without
actually knowing what the solution is.
5.3. Periodic Solutions 131

5.3.1 Closed Solution Curves and Hamiltonians

It is possible to find the equation for the closed curve without too much
trouble if the differential equation comes from Newton’s second law, F =
ma. To explain, if u(t) is the displacement, and F is a function of u, then
F = ma gives us the differential equation mu′′ = F (u). Multiplying this
by the velocity u′ , and remembering that v = u′ , we get that
mvv ′ = F (u)u′ . (5.28)
The key is to observe that the left hand side can be written as
d 1 2

mv .
dt 2
To do the same for the right hand side, let V (u) be such that V ′ (u) =
−F (u). In this case, the right hand side of (5.28) can be written as
dV du d
F (u)u′ = − = − V (u).
du dt dt
What we have done is to rewrite (5.28) as
d 1 
mv 2 + V (u) = 0. (5.29)
dt 2
Integrating this equation,
1
mv 2 + V (u) = c. (5.30)
2
The value of the constant c is determined from the initial condition.
There is a physical interpretation of the equation we have derived that
is worth knowing about. The left hand side of (5.30) is
1
H(u, v) = mv 2 + V (u). (5.31)
2
This function is a Hamiltonian for the differential equation. In this in-
stance it is the total mechanical energy of the system, and it consists of
the sum of the kinetic energy, 21 mv 2 , and a potential energy, V (u). What
we have shown in (5.30) is that the total energy is constant. So, the solu-
tion moves along a constant energy curve determined by the Hamiltonian
and the initial conditions.
Not every forcing function F (u) will result in (5.30) being a closed
curve. Moreover, it is typical that when F (u) is nonlinear, that not all
initial conditions, if any, will yield a closed curve. Examples of these
situations are given below.
Finally, as is often the case in mathematics, it is not recommended that
you memorize the formula given in (5.30). It is better that you remember
how it is derived. Namely, you multiply the second-order equation by the
velocity, and then rewrite the terms as derivatives.
132 Chapter 5. Nonlinear Systems

Example: Mass-Spring Revisited

Starting with the equation

mu′′ + ku = 0,

we multiply by u′ and obtain

mvv ′ + kuu′ = 0.

This can be written as

d 1 1 
mv 2 + ku2 = 0.
dt 2 2
This means that
1 1
mv 2 + ku2 = c,
2 2
where c is an arbitrary constant. Taking, as in the last example, u(0) = 1,
v(0) = 0, m = 1, and k = 4, and substituting these values into the above
equation we find that c = 2. Consequently, the above equation becomes
1
u2 + v 2 = 1. (5.32)
4
This is exactly the same equation (5.27) we derived earlier using the
known solution to the problem. What is significant is that we have found
this curve without first finding the solution of the problem. 

It was mentioned earlier that not every forcing function will result in
a closed curve. For the above mass-spring problem the spring force is
F = −km. This is attractive, in the sense that it pulls the mass back
towards the rest position u = 0. If the force is repelling, so F = km,
then instead of (5.32), you get u2 − 41 v 2 = 1. This is an equation for a
hyperbola, which is not a close curve.

Example: Pendulum
The equation for the angular deflection of a pendulum can be written as

d2 θ
= −α sin θ. (5.33)
dt2
where α = g/ℓ. Introducing the angular velocity v = θ′ , then we obtain
the first-order system

θ′ = v, (5.34)
′
v = −α sin θ. (5.35)
5.3. Periodic Solutions 133

1.6

v-axis
0

-1.6
- /2 - /4 0 /4 /2
-axis

Figure 5.10. Path followed by the solution of the pendulum example.

In this example, assume that α = 4, and that the initial conditions are
θ(0) = π/4 and v(0) = 0. To determine the closed solution curve, we
multiply (5.33) by the velocity v = θ′ , giving us

vv ′ = −4θ′ sin θ.

Writing this as
d 1 2 d

v = 4 cos θ ,
dt 2 dt
and then integrating gives us the equation
1 2
v − 4 cos θ = c.
2
√
With the initial conditions we find that c = −2 2, and so the equation
for the curve takes the form
√
v 2 − 8 cos θ = −4 2. (5.36)

The curve obtained from this equation is shown in Figure 5.10.

The direction of the arrows can be determined ′
′
√ from the v equation
(5.35). Namely, since v (0) = −α sin(θ(0)) = −2 2, and this is negative,
then v must decrease as it leaves the initial point. Therefore, the direction
of motion is clockwise around the curve.
It is possible to determine various properties of the solution from
(5.36). For example, the maximum velocity vM occurs when v ′ = 0.
Since v ′ = −4 sin θ, then from Figure 5.10 it is apparent √ that the only
2
solutionpis θ = 0. In this case, from (5.36), v = 4(2 − 2). Therefore,
√
vM = 2 2 − 2.
Finally, to illustrate the periodicity of the individual components of
the solution, both θ and v are plotted in Figure 5.11 as functions of t.
134 Chapter 5. Nonlinear Systems

v
M

/4

0
v

- /4

-v
M
0 2 4 6 8 10
t-axis

Figure 5.11. Solution curves for θ(t) and v(t) for the pendulum solution
shown in Figure 5.10.

An interesting question is whether it is possible to determine the period

of these functions without knowing the solution. It is, and how this is
possible will be explained in the next section 

Example: Librating versus Circulating Motion

For a pendulum, if the initial velocity is large enough, then the mass will
go all the way around, pass through θ = π (or, θ = −π) and return to
where it started. It will continue to do this indefinitely. This motion is
periodic, but it does not satisfy the definition of a periodic solution given
in (5.22). In mechanics it is called a circulating, or rotating, motion. In
contrast, the tick-toc type of periodic motion considered in the previous
example is referred to as libration.
The integral curves for the pendulum are shown in Figure 5.12. The
closed, solid blue, curves correspond to the periodic solutions discussed
earlier. The dashed curves are some of the possible circulating solutions.
On these curves, the angular coordinate θ increases monotonically if v > 0,

0
v

-6
-3 -2 - 0 2 3

Figure 5.12. Phase portrait for the pendulum equations (5.34), (5.35), when α = 4.
5.3. Periodic Solutions 135

and decreases monotonically if v < 0. In the physical plane this corre-

sponds to the mass continually making complete circuits around the pivot
point (i.e., it is making a circulating motion).
The red curves in Figure 5.12 form what is known as the separatrix
for the pendulum. If you start at a point on the separatrix, the solution
will approach the vertical, unstable, steady state. 

5.3.2 Finding the Period

Once the closed curve formed by the periodic solution is known, it is
possible to find the period. As usual, it is easiest to explained how this
is done using examples.

Example: Mass-Spring
The equation
√ for the curve is given in (5.32). Solving this for v yields
v = ±2 1 − u2 . Which sign you use depends on what part of the curve
you are considering. In Figure 5.9 the two u intercepts are u = ±1. So, for
the lower part
√ of the curve connecting (1, 0) to (−1, 0), v is negative, and
so v = −2 1 − u2 . Since v = u′ , then we have the first-order differential
equation
du p
= −2 1 − u2 .
dt
This equation is separable, which yields
du
Z Z
√ = −2dt.
1 − u2
Carrying out the integrations,
arcsin(u) = −2t + c.
Given that u = 1 at t = 0, then c = π/2.
To determine the period, we solve the above equation for t to obtain
1π 
t= − arcsin(u) .
2 2
It is now possible to determine how long it takes for the solution to move
along the lower half of the curve and arrive at (−1, 0). Namely, letting
u = −1 in the above equation we get that
1π  π
t= − arcsin(−1) = .
2 2 2
To compute the time to transverse the upper part of the curve, you can
either use the separation of variables approach or you can use the sym-
metry of the solution curve. Both yield the result that the time is π/2.
Therefore, the period is the sum, which means that T = π. This agrees
with what we found earlier using the exact solution to the problem. 
136 Chapter 5. Nonlinear Systems

Example: Pendulum
The equation for the curve is given in (5.36). Solving this for v yields
p √
v = ±2 2 cos θ − 2. The lower part of the solution curve, shown in
Figure p
5.10, goes from (π/4, 0) to (−π/4, 0). On this part of the curve
√
v = −2 2 cos θ − 2, which gives us the first-order differential equation
√
q
dθ
= −2 2 cos θ − 2 .
dt
This equation is separable, which yields
dθ
Z
p √ = −2t + c.
2 cos θ − 2
In anticipation of imposing the initial condition, the above integral is
written as Z θ
dr
p √ = −2t + c.
θ0 2 cos r − 2
Now, given that θ(0) = π/4, then θ0 = π/4 and c = 0. The above
equation then takes the form
Z θ
dr
p √ = −2t.
π/4 2 cos r − 2
The time to reach θ = −π/4 is therefore

1 −π/4 dr
Z
t=− p √
2 π/4 2 cos r − 2
1 π/4 dr
Z
= p √ .
2 −π/4 2 cos r − 2

Using the separation of variables approach, or using the symmetry of the

solution curve, the time to transverse the upper part of the curve is the
same as the above value. Therefore, the period T for the pendulum is
Z π/4
dr
T = p √ . (5.37)
−π/4 2 cos r − 2
So, we have a formula for the period that does not require knowing the
solution. The complication is that it is an improper integral, of the type
often referred to in a calculus textbook as “Type II,” which means the
integrand becomes infinite at the endpoints. It is not possible to carry out
the integration in terms of elementary functions, but it is relatively simple
to evaluate it using a computer. Doing so, one finds that T = 3.267 . . ..

Exercises 137

We have been able to determine a great deal about the properties

of a periodic solution, without actually knowing what the solution is.
As stated earlier, this is significant as most of the nonlinear problems
that give rise to a periodic solution can not be solved using elementary
functions. Consequently, they are almost always solved numerically. Our
results complement what can be learned numerically, as we have been able
to derive analytical formulas for the period, the closed curve, and other
components of the solution. This makes it much easier to determine
how the solution changes when the initial conditions, or the parameters
appearing in the equations, are changed.

Exercises
1. Find a Hamiltonian function H(u, v) for each of the following:

a) 2u′′ + 3e2u − 3 = 0 c) 5u′′ + 7u + 6u9 = 0

u √
b) u′′ + =0 d) u′′ + 5u3 1 + u2 = 0
1 + 5u2
2. This problem considers periodic, and non-periodic, solutions of y′ =
f (y).
a) Explain why any steady state is a periodic solution of this equation.
b) Suppose that yb , given below, is a solution. Is it a periodic solution?
! !
sin t sin t
yb = yc =
sin(3t) sin(πt)

c) Suppose that yc , given above, is a solution. Is it a periodic solution?

d) Show that if u(t) is periodic with period T , then v(t) = u′ (t) is also
periodic with period T .
e) Suppose that v = u′ . Give an example where v(t) is periodic, but
u(t) is not periodic.
3. In this problem assume that the curve coming from (5.30) has the form
shown in Figure 5.13.
a) Explain why the maximum, and minimum, velocities occur when
u = us , where us is a steady state value.
b) Use the fact that there are two u-intercepts to explain why it is not
possible that V (u) = u3 .
4. The problem concerns a Duffing oscillator, and the differential equation
is u′′ +u+u3 = 0. Assume the initial conditions are u(0) = 1 and v(0) =
0. This equation comes from a mass-spring system, as shown in Figure
3.2, where the restoring force of the spring is nonlinear (specifically,
cubic) rather than the linear form assumed using Hooke’s law.
138 Chapter 5. Nonlinear Systems

v-axis
0

-1 1
u-axis

Figure 5.13. Path followed by the solution for Exercise 3.

a) The path followed by the solution is shown in Figure 5.14. Find the
equation for this closed curve.
b) Find the steady-state, show that it is not on the curve you found
in part (a).
c) Draw arrows on the curve indicating the direction of motion. Make
sure to explain how you determine this.
d) What is the maximum velocity?
e) What is the minimum displacement?
f) Find a formula, similar to the one in (5.37), for the period.
5. The problem concerns what is known as a Morse oscillator, and the
differential equation is

u′′ + 2 1 − e−u e−u = 0.

Assume the initial conditions are u(0) = 1 and v(0) = 0. This equation
arises when studying the vibrational energy of a diatomic molecule.
a) The path followed by the solution is shown in Figure 5.15. Find the
equation for this closed curve.
b) Find the steady-state, show that it is not on the curve you found
in part (a).
c) Draw arrows on the curve indicating the direction of motion. Make
sure to explain how you determine this.
v-axis

0
u-axis

Figure 5.14. Path followed by the solution of the Duffing oscillator in Exercise 4.
5.4. Motion in a Central Force Field 139

v-axis
0

0
u-axis

Figure 5.15. Path followed by the solution of the Morse oscillator in Exercise 5.

d) What is the maximum velocity?

e) What is the minimum displacement?
f) Find a formula, similar to the one in (5.37), for the period.
6. This problem concerns the generalization of the Hamiltonian to the
general system in (5.4), (5.5). Assume that there is a function H(u, v)
so that
∂H ∂H
= f (u, v) and = −g(u, v). (5.38)
∂v ∂u
Also, assume that f and g are smooth functions of u and v.
a) Explain why this requires that fu = −gv . If this holds then (5.4),
(5.5) is said to be a Hamiltonian system.
b) Show that the H(u, v) given in (5.31) satisfies (5.38).
d
c) Use (5.38) to show that dt H = 0. So, H(u, v) = c, where c is a
constant, and in this sense H(u, v) is a Hamiltonian for the system.
d) Find a Hamiltonian for u′ = v 2 − u, v ′ = v − 2u.
e) Under what condition is the linear system y′ = Ay, where
!
a b
A= ,
c d

a Hamiltonian system? Assuming this holds, find H(u, v).

5.4 Motion in a Central Force Field

The problem of interest concerns the motion in three dimensions of a
particle that is subjected to a radial force F. The specific assumption is
that
1
F = f (r)x, (5.39)
r
where x(t) is the position of the particle and r = ||x||. From Newton’s
second law, the resulting differential equation is
1
mx′′ = f (r)x, (5.40)
r
140 Chapter 5. Nonlinear Systems

Figure 5.16. A particle, the red dot, orbits a particle located at the origin.
The orbit curve lies in a plane containing the origin and has normal n, where n is
parallel to p = x0 × v0 .

where m is the mass of the particle. As for the initial conditions, it

is assumed that the initial position x(0) = x0 and the initial velocity
x′ (0) = v0 are given. To avoid some uninteresting situations, it is assumed
that x0 × v0 6= 0.
The force F can be thought of as coming from the interaction with a
particle located at the center. For example, if the force is gravity, then
f (r) = −k/r2 , where k = GM m. In contrast, if the particles are charged
and the force is electrostatic, then f (r) = −k/r2 , where k = −qQ/4πε0 .
The definition of the various constants making up k is not important here,
other than to know that it is possible for k to be positive or negative. In
particular, it is positive for a gravitational force, and for an electrostatic
force if the charges of the particles are opposite. It is negative for an
electrostatic force if the charges of the particles are the same.
The solution of (5.39) can be shown to lie in a plane that has a normal
vector n that is parallel to p = x0 × v0 (see Exercise 6). We will orient
the coordinate system so the z-axis is in the n direction, which means
that the solution of (5.40) is confined to the x,y-plane. To take advantage
of this, we will use polar coordinates and write x(t) = r(t) cos θ(t) and
y(t) = r(t) sin θ(t). After some routine change of variables calculations
one finds that (5.40) reduces to

m r′′ − r(θ′ )2 = f (r),

 
(5.41)
d 2 ′ 
r (θ ) = 0. (5.42)
dt
The last equation gives us that r2 θ′ = p, where p is a constant, and this
means that the first equation reduces to

mp2
mr′′ = f (r) + . (5.43)
r3
This is a force balance equation, where f (r) is the force introduced earlier
and mp2 /r3 is an outward directed force due to angular momentum.
5.4. Motion in a Central Force Field 141

The second-order differential equation (5.43) can be written as a first-

order nonlinear system by letting v = r′ , giving

r′ = v, (5.44)
1 p2
v′ = f (r) + 3 . (5.45)
m r
It is worth knowing that in the derivation of (5.43), it is found that
p = ||x0 × v0 ||. So, p is a positive constant that is known from the initial
conditions.

5.4.1 Steady States

The steady states, if there are any, satisfy v = 0 and r3 f (r) + mp2 = 0.
Assuming that f (r) = −k/r2 , then to be a steady state it is required that
kr = mp2 . This means we need k > 0, and the resulting steady state is
r = rs , where
mp2
rs = .
k
Also, since r2 θ′ = p, then θ = ωt + θ0 , where ω = k 2 /m2 p3 . The corre-
sponding solution is a circular orbit in the x,y-plane, with radius r = rs
and period 2π/ω.
To check the stability, note that
! !
0 1 0 1
J = 2k 2 = .
mr3
− 3 pr4 0 −p2 /rs4 0
s s

From this one finds that the eigenvalues are ±ip/rs2 , which means that the
stability of the steady state is indeterminate using the Linearized Stability
Theorem.

5.4.2 Periodic Orbit

The next question is whether the solution is periodic. Said another way,
we would like to know if the particle orbits the particle that is located
at the origin. To find the closed curve formed by the solution, if there is
one, we multiply (5.43) by r′ . From this, and remembering that we have
taken f (r) = −k/r2 , it is found that

1 mp2 k
mv 2 + − = c, (5.46)
2 2r2 r
where c is a constant determined by the initial conditions. Completing
the square, we get that
1 1 2
v 2 + p2 − = c20 , (5.47)
r rs
142 Chapter 5. Nonlinear Systems

v-axis

v-axis
0 0

u- 0 u 0 u- u
+ +
u-axis u-axis

Figure 5.17. Two possible elliptical curves coming from (5.48). The u-
intercepts for each ellipse are u− and u+ .

 2
where c20 = v02 + p2 1/r0 − 1/rs , r(0) = r0 , and v(0) = v0 .
To answer the question about a periodic orbit, it will make things
easier if we let u = 1/r. So, (5.47) takes the form
 2
v 2 + p2 u − us = c20 , (5.48)

where us = 1/rs . This is an equation for an ellipse in the u,v-plane with

center (u, v) = (us , 0). Two representative elliptical paths obtained from
this equation are shown in Figure 5.17. Since u = 1/r, then u must
be positive. This means that the dashed portion of the ellipse on the
left is not possible physically. To determine whether the ellipse has only
positive values, we can use the u intercepts. Setting v = 0 in (5.48) yields
u± = us ± c0 /p. As shown in Exercise 7, to have u− > 0 it is required
that k > mr(0)3 [θ′ (0)]2 . Therefore, as long as the initial angular velocity
θ′ (0) is not too large, the particle will orbit the particle at the origin.
To demonstrate what a solution curve looks like, the numerical solu-
tion of the central force problem in (5.43) is shown in Figure 5.18. The
orbital path in the r,v-plane is on the left. The physical path, in the
x,y-plane, is shown on the right.
The question arises as what happens when you get an ellipse like the
one on the left in Figure 5.17. Irrespective of which point you start at

1.5
0.7
v-axis

y-axis

0
0

-1.5 -0.6
0 1 2 0 1 2
r-axis x-axis

Figure 5.18. Numerical solution of (5.43) in the case of when the solution is
periodic. The initial position, and direction of motion, are shown on each curve. The
two time points used to place the direction arrows on the left are the same time points
used on the right.
Exercises 143

on the solid curve, and no matter which direction you go on the curve, u
approaches zero. In other words, r → ∞. Physically, what is happening
is that the angular momentum is so large that an orbit is not possible,
and the particle simply escapes whatever hold the particle at the origin
might have on it. It is also evident from Figure 5.17, contrary to what
is often shown in cartoons, that the particle does not make several orbits
around the origin before escaping. In fact, the particle is incapable of
making even one complete orbit.

Exercises
1. Suppose the law of gravity results in f (r) = −k/r3 , where k > 0. You
can assume that k 6= mp2 .
a) Are there any steady state solutions? If so, check on their stability.
b) Assuming there is a periodic solution, determine its equation in the
u,v-plane.
c) Use your result from part (b) to explain why there is no periodic
solution of this problem.
2. Suppose the law of gravity results in f (r) = −kr, where k > 0. Note
that this is assuming that gravity acts like an elastic spring.
a) Are there any steady state solutions? If so, check on their stability.
b) Assuming there is a periodic solution, determine its equation in the
r,v-plane.
c) The solution curve is shown in Figure 5.19 in the case of when
r(0) = r0 and v(0) = 0. Show that the second r intercept is at
r0 (rs /r0 )2 , where rs is the steady state you found in part (a).
d) Where is the steady state located in Figure 5.19?
3. This problem concerns the solution shown in Figure 5.18.
a) In Figure 5.18(left), where is rs located?
b) In Figure 5.18(right), sketch in the circular orbit derived in Section
5.4.1.
v-axis

| |
0
r-axis

Figure 5.19. Solution curve for the problem in Exercise 2.

144 Chapter 5. Nonlinear Systems

4. What initial conditions correspond to someone throwing a baseball so

that it encircles the Earth at a constant height, and then returns to
the person who threw it? Some of the results from Exercise 7 might
be useful here. Assume the Earth is a smooth sphere with radius R.
5. This exercise explores the usefulness of making the change of variables
from r, t to u, τ , where u(τ ) = 1/r and τ = θ(t). This is an approach
often used in physics textbooks.
a) Show that r′ (t) = −pu′ (τ ), and r′′ (t) = −p2 u2 u′′ (τ ).
b) The mathematical requirement for the change of variables to be
valid is that θ(t) is a strictly monotonic function of t. Explain why
this holds in this problem.
c) Using the results from part (a), show that (5.43) takes the form
1 1
u′′ + u = − 2 2 f .
mp u u
d) Assuming that f (r) = −k/r2 , find the general solution of the result-
ing differential equation in part (c). In doing this, use (3.23) when
writing down the general solution of the associated homogeneous
equation. Also, what is the resulting formula for r?
e) Use the results from Exercise 7 to show that u(0) = 1/x0 and
u′ (0) = −x′0 /(x0 y0′ ). Use these to find the two arbitrary constants
in your solution in (d).
6. Let p = x × x′ . In this exercise you will likely need to review the
properties of the cross product you learned in calculus.
a) Show that p′ = 0. This means that p is a constant vector, and
so, from the initial conditions, p = x0 × v0 . It is assumed that
x0 × v0 6= 0.
b) Explain why p · x = 0 and p · x′ = 0. Why does this mean that
x and x′ are in the plane that is perpendicular to p, and which
contains the origin?
c) Assuming x = (r(t) cos θ(t), r(t) sin θ(t), 0), show that p = r2 θ′ k,
where k is the unit vector pointing in the positive z-direction.
d) The plane has normal p as well as normal −p. Which one is used
when orientating the positive z-axis in such a way that r2 θ′ > 0?
7. This problem determines how the initial conditions for (5.39) con-
tribute to the reduced problem for the orbit. Assume that x0 =
(x0 , 0, 0)T and v0 = (x′0 , y0′ , 0)T , where x0 , x′0 , and y0′ are given with
x0 and y0′ both positive. The superscript T indicates transpose. Also,
assume that f (r) = −k/r2 and θ(0) = 0.
a) Show that p = x0 y0′ .
b) Show that the initial conditions for (5.43) are r(0) = x0 , r′ (0) = x′0 .
c) Show that u− > 0 reduces to the requirement that k > mx0 (y0′ )2 .
 Chapter 6

Laplace Transform

We have found that to solve y ′′ + by ′ + cy = 0 you assume that y = ert ,

and for x′ = Ax you assume that x = aert . What is notable here is the
exponential dependence of the solution on t. It is possible to extend this
assumption in such a way that it is possible to solve a wide variety of
more complicated problems, such as those involving partial differential
equations. The extension we are going to consider is called the Laplace
transform.
It is recommended that if you are a bit fuzzy on integration by parts,
or partial fractions, that you spend some time reviewing those integration
methods as one or both are used in many of the examples and exercises
in this chapter.

6.1 Definition
The generalization we are interested in called the Laplace transform, and
its definition is given next.

Laplace Transform. Given a function y(t), for 0 ≤ t < ∞, its Laplace

transform Y (s) is defined as
Z ∞
Y (s) ≡ y(t)e−st dt. (6.1)
0

It will be useful to have a more compact notation for the integral in this
expression, and this will be done by writing the above formula as

Y (s) ≡ L(y). (6.2)

Introduction to Differential Equations, M. H. Holmes, 2020

145
146 Chapter 6. Laplace Transform

The Laplace variable s is analogous to the r used in the assumption y = ert

or x = aert . Also, although it is not apparent from the definition, it is
important to know that the variable s is complex-valued.

6.1.1 Requirements
There are a couple of ways to find the Laplace transform of a function.
One is to carry out the integration in (6.1), just as you did in calculus.
Much of the material in this section concerns the mathematical require-
ments needed to do this. However, it is also possible to find a Laplace
transform by looking it up in a table, such as the one in Table 6.1 (on
page 151). As demonstrated in Example 3 (version 2), this is very easy
to do. Most, but not all, of the transforms in this chapter can be done
using the given table (along with the formula in Exercise 4).
For the improper integral in (6.1) to exist, a condition must be im-
posed on the complex variable s. To explain, if y(t) = e3t , then using the
definition of an improper integral and (6.1)
Z T Z T
3t −st
Y (s) = lim e e dt = lim e(3−s)t dt (6.3)
T →∞ 0 T →∞ 0
 
1 (3−s)T 1
= lim e − . (6.4)
T →∞ 3−s 3−s
Clearly, we need s 6= 3. As for the limit, it is useful to know that, given
a nonzero complex number z,
(
zT 0 if Re(z) < 0,
lim e = (6.5)
T →∞ does not exist if Re(z) ≥ 0.
The proof of this comes directly from Euler’s formula, as expressed in
(3.15). For (6.4), z = 3 − s and this means that for the limit to exist we
need Re(3 − s) < 0, or equivalently, we need Re(s) > 3. In this case,
1
Y (s) = .
s−3
The requirement that Re(s) > 3 gives rise to what is known as the half-
plane of convergence for the Laplace transform.
The second mathematical requirement concerns the smoothness of
y(t). For the problems considered in this textbook, it is enough to as-
sume that y(t) is continuous for 0 ≤ t < ∞, except possibly for jump
discontinuities. What a jump discontinuity means is that y(t) is not
continuous at the point, but the limits of y from the left and right are
defined and finite (the two limits do not need to be equal). A simple
example, with a jump discontinuity at t = 2, is
(
3 if 0 ≤ t ≤ 2,
y(t) = (6.6)
−1 if 2 < t.
6.1. Definition 147

The specific requirement for the Laplace transform is that over any inter-
val 0 ≤ t ≤ T , y(t) is continuous except for possibly a finite number of
jump discontinuities. In this case, y(t) is said to be piecewise continu-
ous for t ≥ 0.
The final requirement on y(t) is to guarantee that the improper inte-
gral in (6.1) converges. Specifically, it is required that there is a constant
α so that
lim y(t)eαt = 0.
t→∞

If this holds, then y(t) is said to have exponential order. As examples, any
polynomial function in t, any linear combination of sin(ωt) and cos(ωt),
and any linear combination of terms of the form eωt have exponential
2 3
order. On the other hand, et and et do not. Throughout this chapter,
whenever taking the Laplace transform, it is assumed that the function
has exponential order.
In the above example it is stated that Re(s) > 3. This can not be
written as s > 3. The reason is that the usual definition of inequality can
not be used with complex numbers. For example, the rules of inequality
require that s2 ≥ 0. So, taking s = i, you end up concluding that −1 ≥ 0.
In fact, it is possible to prove that the complex numbers cannot be made
an ordered field no matter how you define the rule used for inequality.

6.1.2 Examples
With the technical details out of the way, we consider a few examples. As
you will see, finding the Laplace transform of a function provides ample
opportunity to practice using integration by parts. Also, in what follows
the improper integral will be treated as a definite integral, with the upper
endpoint being t = ∞ (see Example 1 below). It is understood that the
evaluation at the upper endpoint involves a limit, as expressed in (6.3).

Example 1: If y(t) = sin 2t, find Y (s).

Answer: Using (6.1), and integration by parts (with u = e−st and
dv = sin 2t dt),
Z ∞
Y (s) = sin(2t)e−st dt
0

1 s ∞
Z
∞
−st
= − cos(2t)e − cos(2t)e−st dt
2 t=0 2 0
1 s ∞
Z
= − cos(2t)e−st dt.
2 2 0

To guarantee that cos(2t)e−st has a finite limit as t → ∞, it has

been assumed that Re(s) > 0. Using integration by parts again,
148 Chapter 6. Laplace Transform

you find that

1 s2
Y (s) = − Y (s).
2 4
Solving for Y , we get that Y = 2/(s2 + 4). Using the L(y) notation,
we have that
2
L(sin 2t) = 2 . 
s +4
Example 2: If y(t) is given in (6.6), find Y (s).
Answer: Using the additive property of integrals,
Z ∞
Y (s) = y(t)e−st dt
0
Z 2 Z ∞
−st
= y(t)e dt + y(t)e−st dt.
0 2

Consequently, from (6.6),

Z 2 Z ∞
−st
Y (s) = 3e dt − e−st dt
0 2

3 2 1 ∞
= − e−st + e−st
s t=0 s t=2

4 3
= − e−2s + .
s s
To guarantee that 1s e−st has a defined limit as t → ∞, it has been
assumed that Re(s) > 0. 

In the above two examples, the condition on s so that Y (s) is defined

was stated explicitly. In the remainder of the chapter this will not be
done, and it is assumed that the condition is obvious from the derivation.

Example 3 (version 1): If y(t) = 3t − sin 2t, find Y (s).

Answer: Using integration by parts, and the result from Example
1,
Z ∞ Z ∞
Y (s) = 3 te−st dt − sin 2te−st dt (6.7)
0 0
∞
3t −st 3 2
Z
∞
=− e + e−st dt −
s t=0 s 0 s2 +4
3 2
= − . 
s2 s2 + 4
Exercises 149

Linear Operator

It states in (6.7) that L(3t − sin 2t) = 3L(t) − L(sin 2t). This is an
illustration of an essential property of the Laplace transform. Namely, if
c1 and c2 are constants, then


L c1 y1 + c2 y2 = c1 L(y1 ) + c2 L(y2 ). (6.8)
Another way to write this is to let y(t) = c1 y1 (t) + c2 y2 (t), in which case
Y (s) = c1 Y1 (s) + c2 Y2 (s), (6.9)
where Y1 and Y2 are the Laplace transforms for y1 and y2 , respectively.
Because the Laplace transform has this property, it is said to be a linear
operator. The usefulness of the linearity of the Laplace transform is why
it is listed first in Table 6.1.
It is worth pointing out that you know several other linear operators.
One is a matrix, because it satisfies A(c1y1 + c2 y2 ) = c1 Ay  1 + c2dAy2 . A
d
second is differentiation, as it satisfies dt c1 y1 (t) + c2 y2 (t) = c1 dt y1 (t) +
d
c2 dt y2 (t). You are probably wondering if it’s possible for mathematicians,
or even engineers, to write entire textbooks on linear operators. Well, a
Google book search will answer this question.

Using a Table

Table 6.1, and specifically Properties 7-10 in the table, make it easier to
find some of the more common Laplace transforms. You simply find the
function y(t) in the third column, and then its Laplace transform Y (s) is
in the second column. The next example illustrates how this is done.
Example 3 (version 2): If y(t) = 3t − sin 2t, find Y (s).
Answer: Using linearity, as expressed in (6.8),


Y (s) = L 3t − sin 2t = 3L(t) − L(sin 2t).
From Table 6.1, using Property 8 (with n = 1 and a = 0), and
Property 9 (with a = 0 and ω = 2),
3 2
Y (s) = 2
− 2 . 
s s +4
It is possible to extend the usefulness of Table 6.1 by using some of the
properties of a Laplace transform. Exercise 4 considers one of particular
note.

Exercises
1. Find the Laplace transform of the following functions.
150 Chapter 6. Laplace Transform

a) y = −e5t d) y = e−2t − e7t g) y = te−t

b) y = 3 + 4t e) y = 4t2 h) y = 2 sin(3πt + 4)
c) y = 2t + 4e−t f) y = (t − 3)2 i) y = e2t + 4 cos(2t)

2. Find the Laplace transform of the following functions.

( (
0 if 0 ≤ t ≤ 2, 3 if 0 ≤ t ≤ 1,
a) y(t) = d) y(t) =
5 if 2 < t e−t if 1 < t
( (
t if 0 ≤ t ≤ 2, 5 − t if 0 ≤ t ≤ 3,
b) y(t) = e) y(t) =
2 if 2 < t t − 1 if 3 < t
 
 −1 if 0 ≤ t ≤ 1,
  0 if 0 ≤ t ≤ 12,

c) y(t) = 1 if 1 < t < 2, f) y(t) = 2 if 12 < t < 15,
 
0 if 2 ≤ t 0 if 15 ≤ t
 

3. One way to avoid using integration by parts is
to use the formulas
cos x = 21 eix + e−ix and sin x = 2i1
eix − e−ix (see Section 3.4.1).

Use these to find the Laplace transform of y(t).

a) y(t) = cos(3t) c) y(t) = e−t cos(πt)

b) y(t) = 4 sin(7t) d) y(t) = cos(t) sin(2t)

4. The exercise explores the usefulness of the formula

Z ∞ Z ∞
−st d
ty(t)e dt = − y(t)e−st dt.
0 ds 0

Use this, and Table 6.1, to find the Laplace transform of the following
functions:

a) t sin(3t) b) 6t cos(7t) c) t2 cos(t) d) te−2t sin(5t)

6.2 Inverse Laplace Transform

As will be seen when we get around to solving differential equations, we
will use the Laplace transform to change the problem from solving for y
to solving for Y . It is actually fairly easy to do this. Once Y is known,
it is then necessary to determine y. This requires us to know how to find
the inverse Laplace transform.
Using the L(y) notation, the inverse Laplace transform is written as
L−1 (Y ). As an example, earlier we found that
2
L(sin 2t) = .
s2 + 4
6.2. Inverse Laplace Transform 151

Y (s) = L(y) y(t) = L−1 (Y )

1. aY (s) + bV (s) ay(t) + bv(t)

Z t
2. V (s)Y (s) v(t − r)y(r)dr
0

3. sY (s) y ′ (t) + y(0)

Z t
1
4. s Y (s) y(r)dr
0

5. e−as Y (s) y(t − a)H(t − a) for a > 0

6. Y (s + a) e−at y(t)

1 −as
7. se H(t − a) for a > 0

n!
8. (s+a)n+1 tn e−at for n = 0, 1, 2, 3, . . .

ω
9. (s+a)2 +ω 2 e−at sin(ωt)

s+a
10. (s+a)2 +ω 2 e−at cos(ωt)
h i
c + (ac + d)t eat if s1 = s2 = a
cs+d 1
h
at − (bc + d)ebt
i s1 = a
11. (s−s )(s−s ) a−b (ac + d)e if s2 = b
1 2

s = a + ib
h i
eat c cos(bt) + ac+db sin(bt) if s1 = a − ib
2

p(s)
12. (s−s )(s−s )···(s−s ) see (6.16)
1 2 n

13. e−as δ(t − a) for a > 0

14.
1 −as 1 n−1 e−b(t−a) H(t−a) for a > 0,
(s+b)n e (n−1)! (t−a) n = 1, 2, 3, . . .
Z t−a
1 −as
15. se Y (s) H(t − a) y(r)dr for a > 0
0
Table 6.1. Laplace and inverse Laplace transforms. The function H(x) is
defined in (6.13), and δ(t) is defined in Section 6.5.1. Also, recall that 0! = 1, and if
t > 0, then t0 = 1.
152 Chapter 6. Laplace Transform

The inverse is therefore

 2 
L−1 = sin 2t .
s2 + 4

The caveat here is that if Y = L(y), it is not always true that y = L−1 (Y ).
It is true for the above example, and this is because the original function
y(t) is continuous. What happens when y(t) has a jump discontinuity will
be discussed later.
In Section 6.4, the first differential equation we will solve using a
1 1
Laplace transform is y ′ + 3y = e2t . We will find that Y = s+3 (2 − 2−s ),
−1
and this will mean that to find y we will need to determine L (Y ). There
is a general formula for the inverse Laplace transform, which involves
a line integral in the complex plane. Although this can provide some
entertaining mathematical challenges, most find the inverse transform by
using tables. Table 6.1 is an example, and it is the one used in this text.
Note that the first six entries are general properties for the transform.
The first one listed is the linearity property, as given in (6.8). Writing it
in terms of the inverse transform, we have that

L−1 c1 Y1 + c2 Y2 = c1 L−1 (Y1 ) + c2 L−1 (Y2 ).

This is used for most of the examples in this chapter.

As demonstrated in the next example, when using a table, finding an
inverse transform can be very easy.

3 7s
Example: If Y (s) = s2 − s2 +25 , find y(t).
Answer: Using the linearity property,
3 7s 
L−1 (Y ) = L−1 −
s2 s2 + 25
1  s 
= 3L−1 2 − 7L−1 2 .
s s + 25

From Property 8 in Table 6.1, with n = 1 and a = 0,

1
L−1 = t.
s2
Similarly, from Property 10, with a = 0, and ω = 5,
 s 
L−1 2 = cos(5t).
s + 25

Therefore,
y(t) = 3t − 7 cos(5t). 
6.2. Inverse Laplace Transform 153

6.2.1 Jump Discontinuities

At points t where the original function y(t) has a jump discontinuity, then
L−1 (Y ) equals the average in the jump in y. The formula is, for t > 0,
1 +
L−1 (Y ) = y(t ) + y(t− ) .

(6.10)
2
To illustrate, earlier we found that if
(
3 if 0 ≤ t ≤ 2,
y(t) = (6.11)
−1 if 2 < t,

then
4 3
Y (s) = − e−2s + .
s s
Except
 for t = 2, L−1 (Y ) = y. At t = 2, the average in the jump in y(t)
is 2 y(2 ) + y(2− ) = 21 (3 − 1) = 1. Therefore, the inverse transform is
1 +



 3 if 0 ≤ t < 2,
−1
L (Y ) = 1 if t = 2, (6.12)

−1 if 2 < t.


It is convenient to use what is called the Heaviside step function

H(x) when jumps occur. This is defined as

 0
 if x < 0,
1
H(x) ≡ 2 if x = 0, (6.13)

1 if 0 < x,


and this is shown in Figure 6.1. Note that this has built into its definition
the value at a jump that is needed for the inverse Laplace transform.
To rewrite (6.12) using H(x), since L−1 (Y ) involves a jump of −4 at
t = 2, then L−1 (Y ) = 3 − 4H(t − 2).

1
H(x)

0.5

0
-1 0 1

Figure 6.1. Heaviside step function H(x) as defined in (6.13).

154 Chapter 6. Laplace Transform

2 5 6
Example: If Y = s + s e−3s − s e−4s , find and then sketch y.
Answer: Using the linearity property,
2 5 6 
L−1 (Y ) = L−1 + e−3s − e−4s
s s s
1 1  1 
= 2L−1 + 5L−1 e−3s − 6L−1 e−4s .
s s s
From Property 8 in Table 6.1, with n = 0 and a = 0,
1
L−1 = 1.
s
From Property 7, with a = 3 and a = 4,
1  1 
L−1 e−3s = H(t − 3), and L−1 e−4s = H(t − 4).
s s
Therefore,
y(t) = 2 + 5H(t − 3) − 6H(t − 4). (6.14)

So, the solution starts out at y = 2, it has a jump of 5 at t = 3, so

y = 7, and then it has another jump of −6 at t = 4, so y = 1. At
the jumps, y(3) = 21 (2 + 7) = 29 , and y(4) = 12 (7 + 1) = 4. The plot
of this function is given in Figure 6.2. 

For those who are picky about doing things correctly, there is a mild
case of notation abuse in the last example. Because the function y(t)
has a jump, and we only know its Laplace transform, it is not possible
to determine the value of y(t) at the jump. The function given in (6.14)
is the answer that is consistent with the formula determined using the
inverse Laplace transform. This situation will arise in this chapter any
time the function y(t) has a jump discontinuity.

6
y(t)

0
0 1 2 3 4 5 6
t

Figure 6.2. The function y(t) given in (6.14).

Exercises 155

6.2.2 Heaviside Expansion Theorem

There are numerous special cases for which formulas can be found for the
inverse Laplace transform. One that deserves to be mentioned is the case
of when Y (s) can be written in the form
p(s)
Y (s) = , (6.15)
(s − s1 )(s − s2 ) · · · (s − sn )
where p(s) is a polynomial of degree less than n. The sj ’s can be complex-
valued but they must be distinct, which means that sj 6= sk if j 6= k. In
this case, using something called residue theory, it can be shown that the
inverse transform L−1 (Y ) is
n
X p(sj ) sj t
y(t) = e , (6.16)
q ′ (sj )
j=1

where q(s) = (s − s1 )(s − s2 ) · · · (s − sn ). The above result is often called

the Heaviside Expansion Theorem.

Example: Find the inverse transform of

2
Y = .
(s2 + 1)(s2 + 4)
Answer: To use the Heaviside Expansion Theorem, note that q(s) =
(s + i)(s − i)(s − 2i)(s + 2i). So, s1 = −i, s2 = i, s3 = 2i, and
s4 = −2i. To compute q ′ (sj ) it makes it a bit easier if you note that
q(s) = (s2 + 1)(s2 + 4) = s4 + 5s2 + 4. So, q ′ (s) = 4s3 + 10s, and
this means that q ′ (−i) = 4i − 10i = −6i, q ′ (i) = −6i, q ′ (2i) = −12i,
and q ′ (−2i) = 12i. With this, from (6.16) we get that
2 2 2 2
y(t) = e s1 t + e s2 t + e s3 t + e s4 t
q ′ (s 1) q ′ (s 2) q ′ (s 3) q ′ (s 4)
2 −it 2 2 2it 2 −2it
= e + eit + e + e
−6i 6i −12i 12i
1 it 1 2it
e − e−it − e − e−2it .



=
3i 6i
Since eiθ − e−iθ = 2i sin θ, then
2 1
y(t) = sin t − sin 2t.  (6.17)
3 3

Exercises
1. Sketch the function for 0 ≤ t ≤ T , and then find its Laplace transform.
156 Chapter 6. Laplace Transform

a) y = H(t − 6), T = 10 d) y = 3H(t − 2) − 4H(t − 5), T = 8

b) y = (t − 1)H(t − 1), T = 3 e) y = 4H(3 − t), T = 5
c) y = [H(t − 2)]2 , T = 4 f) y = 1/(2 + H(t − 1)), T = 3
g) y = H(t) − H(t − 1) + 3H(t − 2) − H(t − 3), T = 5
h) y = H(t) + H(t − 2) − H(4 − t) − H(t − 8), T = 10

2. Find the inverse Laplace transform of the following functions.

2 2 3
a) Y = s2 +9 i) Y = s2 +4 − s2 +9
3 5 3
b) Y = (s+4)2 j) Y = s2 −
s−2
1 s+1
c) Y = s2 +3s−4 k) Y = (s+1)2 +9 e−3s
s+1
d) Y = s2 +2s+5 1 1


l) Y = s2 − s3 e−2s
2s−3
e) Y = s2 −4 1
= s e−s − e−2s + e−3s


m)Y
2s−3
f) Y = s2 +2s+10 n) Y
2 1 4
= s − s2 + s3 − s4
7
s 5s+1
g) Y = (s2 +9)(s2 +16) o) Y = s2 e−5s
s 5s+1
h) Y = (s2 −1)(s2 −2) p) Y = s2 +1 e−6s

3. Suppose that y(t) is periodic with period T > 0. So, y(t + T ) = y(t)
for all t ≥ 0.
a) Show that Z ∞
y(t)e−st dt = e−sT Y (s).
T
R∞ RT R∞
b) Writing 0 y(t)e−st dt = 0 y(t)e−st dt + T y(t)e−st dt, use the re-
sult from part (a) to show that
T
1
Z
L(y) = y(t)e−st dt.
1 − e−sT 0

4. The following functions are periodic with period T . Sketch the function
for 0 ≤ t ≤ 3T , and then use the result of Exercise 3(b) to find the
Laplace transform. Also, provide an explanation for where the name
of the wave comes from.
a) Square wave: T = 2, and y(t) = H(t) − H(t − 1), for 0 ≤ t < 2.
b) Sawtooth wave: T = 1, and y(t) = t, for 0 ≤ t < 1.
c) Triangle wave: T = 2, and y(t) = tH(t) − 2(t − 1)H(t − 1), for
0 ≤ t < 2.
d) Bang-bang wave: T = 2, and y(t) = H(t) − 2H(t − 1), for 0 ≤ t < 2.
6.3. Properties of the Laplace Transform 157

5. The floor function y(t) = ⌊t⌋ is the greatest integer less than or equal
to t. So, ⌊5.3⌋ = 5 and ⌊7.0⌋ = 7.
a) Writing ⌊t⌋ = t − g(t), what are: g(0), g(0.1), g(0.8), and g(1)?
b) Sketch g(t) for 0 ≤ t < 5. Use this to explain why g(t) is periodic.
c) Use the result from Exercise 3(b) to find L(⌊t⌋).
6. It is sometimes necessary to use a power series to determine an inverse
transform. For example, to use the geometric series (1 − z)−1 = 1 −
z + z 2 + · · · to write (1 − e−s )−1 = 1 − e−s + e−2s + · · · . In this problem
you are to use a Maclaurin series to find the inverse Laplace transform.
1
a) Y = s(1−e−s ) , use the geometric series for (1 − z)−1
1
b) Y = √ , use the series for (1 + z)−1/2
s 1+e−s
1p √
c) Y = s 1 + (1/s), use the series for 1 + z

6.3 Properties of the Laplace Transform

What follows is the derivation of Properties 2 and 3 in Table 6.1. They
are important as they will be needed when solving differential equations.

6.3.1 Transformation of Derivatives

One of the hallmarks of the Laplace transform, as with most integral
transforms, is that it converts differentiation into multiplication. To ex-
plain what this means, using integration by parts (with u = e−st and
dv = y ′ (t)dt), we have the following:
Z ∞
′
L(y (t)) = y ′ (t)e−st dt
0
Z ∞
∞
−st
= ye +s ye−st dt
t=0 0

= sL(y) − y(0). (6.18)

This holds assuming that y ′ (t) is piecewise continuous, and y(t) is con-
tinuous, for t ≥ 0.
The above formula can be used to find the transform of higher deriva-
tives. As an example, for the second derivative we have that
L(y ′′ ) = sL(y ′ ) − y ′ (0) = s sL(y) − y(0) − y ′ (0)
 

= s2 L(y) − y ′ (0) − sy(0). (6.19)

The requirement here is that y ′′ (t) is piecewise continuous, and y(t) and
y ′ (t) are continuous, for t ≥ 0. Generalizing this to higher derivatives,
L(y (n) ) = sn L(y) − y (n−1) (0) − sy (n−2) (0) − · · · − sn−1 y(0),
with the corresponding generalization in the continuity requirements.
158 Chapter 6. Laplace Transform

6.3.2 Convolution Theorem

A common integral that arises when solving differential equations is a
convolution integral of the form
Z t
y(t) = g(t − τ )v(τ )dτ . (6.20)
0

Taking the Laplace transform of this equation we obtain

Z ∞Z t
L(y) = g(t − τ )v(τ )e−st dτ dt
0 0
Z ∞Z ∞
= g(t − τ )v(τ )e−st dtdτ.
0 τ

In the last line above, interchanging the order of integration used the fact
that 0 < t < ∞, 0 < τ < t is equivalent to 0 < τ < ∞, τ < t < ∞. Now,
making the change of variables t = r + τ in the inner integral, we get that
Z ∞Z ∞
L(y) = g(r)v(τ )e−s(r+t) drdτ
Z0 ∞ 0 Z ∞ 
−sτ −sr
= v(τ )e g(r)e dr dτ = V (s)G(s).
0 0

Using the inverse transform this can be written as

Z t
−1
L (V (s)G(s)) = g(t − τ )v(τ )dτ . (6.21)
0

This is Property 2, in Table 6.1, and it is known as the convolution the-

orem.

Exercises
1. Find the Laplace transform in terms of Y (s).
a) L y ′ − 4y , where y(0) = 1

b) L 2y ′ + 7y , where y(0) = −2

c) L y ′′ + 5y , where y(0) = 1 and y ′ (0) = −1

d) L y ′′ + 3y ′ − 2y , where y(0) = 1 and y ′ (0) = −3

e) L 4y ′′ + 2y ′ , where y(0) = −1 and y ′ (0) = 1

2. Use the convolution theorem, as given in (6.21), to find the inverse

transform. Note that how you select G and V will determine how
difficult the resulting integral is to carry out.
1 1 1
a) (s−1)(s2 +1) , taking G(s) = s−1 and V (s) = s2 +1
6.4. Solving Differential Equations 159

s 1 s
b) (s2 +1)2 , taking V (s) = s2 +1 and G(s) = s2 +1
5
c) (s+1)(s2 +4)
1
d) (s2 +1)(s2 −1)
1
e) s3 (s2 +1)
3. Explain why Property 4 in Table 6.1 is a special case of the convolution
theorem.

6.4 Solving Differential Equations

The examples to follow illustrate how to use the Laplace transform to
solve a linear initial value problem. As you will see, it is fairly easy to
transform the equation and then solve for Y (s). Most of the work is done
trying to determine the inverse transform to find y(t).

Example 1: Solve y ′ + 3y = e2t , where y(0) = 2.

Answer: The first step is to take the Laplace transform of the dif-
ferential equation, which gives

L(y ′ + 3y) = L(e2t ).

Using the linearity of the transform, and the derivative formula

(6.18),

L(y ′ + 3y) = L(y ′ ) + 3L(y)

= sL(y) − y(0) + 3L(y) (6.22)
= (s + 3)Y (s) − 2.

Also,
∞ ∞
1
Z Z
2t 2t −st
L(e ) = e e dt = e(2−s)t dt = .
0 0 s−2

The transformed problem is therefore (s + 3)Y − 2 = 1/(s − 2), and

from this we get that
1  1 
Y = 2+ .
s+3 s−2
Consequently, using Table 6.1 (Properties 7 and 11),
 1   1 
y = L−1 (Y ) = 2L−1 + L−1
s+3 (s + 3)(s − 2)
1
9 1
= 2e−3t + e2t − e−3t = e−3t + e2t . 
5 5 5
160 Chapter 6. Laplace Transform

Example 2: Solve y ′′ + y ′ − 2y = − sin t, where y(0) = 1 and y ′ (0) = 1.

Answer: Taking the Laplace transform of the differential equation

L(y ′′ + y ′ − 2y) = L(− sin t).

Using the linearity of the transform, and the derivative formulas

(6.18) and (6.19),

L(y ′′ + y ′ − 2y) = L(y ′′ ) + L(y ′ ) − 2L(y)

= s2 Y − y ′ (0) − sy(0) + sY − y(0) − 2Y (6.23)
2
= (s + s − 2)Y − s − 2.

Since, using Property 9 in Table 6.1, L(sin t) = 1/(s2 + 1), then the
transformed problem is
1
(s2 + s − 2)Y − s − 2 = − .
s2 + 1
Solving for Y gives us
1 1
Y = − 2 . (6.24)
s − 1 (s + 1)(s2 + s − 2)

Taking the inverse transform, and using linearity,

 1   1 
y = L−1 − L−1 .
s−1 (s2 + 1)(s2 + s − 2)

Using Table 6.1, Property 8,

 1 
L−1 = et .
s−1
For the other inverse transform, we will use partial fractions. The
assumption is that
1 As + B Cs + D
= 2 + 2
(s2 + 1)(s2 + s − 2) s +1 s +s−2
(A + C)s3 + (A + B + D)s2 + (−2A + B + C)s − 2B + D
= .
(s2 + 1)(s2 + s − 2)

Equating like powers of s in the numerators, we get that A + C = 0,

A + B + D = 0, −2A + B + C = 0, and −2B + D = 1. Solving
these equations one finds that A = −1/10, B = −3/10, C = 1/10,
and D = −2/5. So, using Table 6.1, Properties 10 (for A), 9 (for
B), and 11 (for C and D),
6.4. Solving Differential Equations 161

 1 
−1 As + B
   Cs + D 
L−1 = L + L −1
(s2 + 1)(s2 + s − 2) s2 + 1 s2 + s − 2
1 h i
= A cos t + B sin t + (C + D)et + (2C − D)e−2t .
2
(6.25)
Therefore, the solution is
1 3 1 5
y= cos t + sin t + e−2t + et . 
10 10 15 6

Two comments need to be made about the above examples. First,

both can be solved much easier using the method of undetermined coef-
ficients. The Laplace transform was used to illustrate how it can be used
to solve such problems. Second, the question invariability comes up as to
what is the easiest way to determine the inverse transform. For example,
you can use the Heaviside expansion theorem, the convolution theorem, or
partial fractions to obtain (6.25). There is often no clear answer to which
one to use, and it often depends on what you are the most comfortable
with and what is applicable. As it turns out, except perhaps when taking
a course in differential equations, very few people work out even slightly
complicated inverse transforms by hand. Instead, they either buy a book
of tables, such as Oberhettinger and Badii [1973], or, even more likely,
they use a symbolic computing system like Maple or Mathematica.

6.4.1 The Transfer Function

In engineering, when solving a linear differential equation, it is common
to introduce what is known as the transfer function H(s). To find H(s)
you take the Laplace transform of the differential equation, assuming
the initial conditions are all zero, and then solve for Y (s). This yields
Y (s) = H(s)F (s), where F (s) is the Laplace transform of the forcing
function. In this sense, H(s) is the transfer function from the input F (s)
to the output Y (s). It should be pointed out that the transfer function
is usually denoted as H(s). In this text, H(s) is used to avoid confusing
it with the Heaviside function.
√
Example 1: Find the transfer function for y ′ + 3y = t.
Answer:√ Taking y(0) = 0, then from (6.22), (s + 3)Y = F , where
F = L( t). Consequently, H(s) = 1/(s + 3). 
Example 2: Find the transfer function for y ′′ + y ′ − 2y = ln(1 + t2 ).
Answer: Taking y(0) = 0 and y ′ (0) = 0, then from (6.23), we have
that (s2 + s − 2)Y = F , where F = L(ln(1 + t2 )). Consequently,
H(s) = 1/(s2 + s − 2). 
162 Chapter 6. Laplace Transform

Once you know the transfer function, then a particular solution of the
differential equation can be written down using the convolution theorem.
Namely, using (6.21), a particular solution is
Z t
yp (t) = h(t − τ )f (τ )dτ , (6.26)
0

where h(t) = L−1 (H).

The above solution is useful as it can be used to solve the IVP when
the initial conditions are not zero. This is because the solution can be
written as y(t) = yp (t) + yh (t), where yp (t) is given in (6.26) and yh (t)
is the solution of the associated homogeneous differential equation that
satisfies the nonzero initial conditions. This is similar to the approach
used in Section 3.9.1. Moreover, it is not necessary to use the Laplace
transform to find yh (t). An example of solving an IVP in this way is given
next.
√
Example 3: Solve y ′′ − 2y ′ − 3y = t, where y(0) = 1 and y ′ (0) = −1.
√
Step 1: Find yp . The transfer function for y ′′ − 2y ′ − 3y = t
is H(s) = 1/(s2 − 2s − 3). Using Property 11 from Table 6.1,
h(t) = L−1 (H) = (e3t − e−t )/4. So, from (6.26),
1 t  3(t−τ )
Z
− e−t+τ τ dτ.
√
yp (t) = e
4 0
The integral can not be written in terms of elementary functions,
and so this is the final answer.
Step 2: Find yh . The IVP to solve is y ′′ − 2y ′ − 3y = 0, where
y(0) = 1 and y ′ (0) = −1. Assuming that y = ert , and proceeding
as in Section 3.5, one ends up finding that yh (t) = e−t .
Step 3: The solution is y = yp + yh . In other words,

1 t h 3(t−τ ) i√
Z
−t
y(t) = e + e − e−t+τ τ dτ. 
4 0

6.4.2 Comments and Limitations on Using the Laplace Transform

It is useful to know some of the limitations on using the Laplace transform
to solve a differential equation. First, the differential equation must be
linear. As the examples illustrate, the Laplace transform can be used
irrespective of the order of the equation. It can also be used to solve partial
differential equations, delay equations, and integral equations. However,
in all cases, the equations are linear.
A second limitation is that the differential equation should have con-
stant coefficients. For example, the Laplace transform will not be suc-
cessful when trying to solve y ′ + et y = 0 or y ′′ + (1 + t)2 y ′ + 5y = 0.
Exercises 163

Occasionally you will come across an equation with non-constant coeffi-

cients that can be solved using a Laplace transform, and an example is
Airy’s equation y ′′ + ty = 0. You might try finding the Laplace transform
of this equation to see why the coefficients are “just right” so that the
method works.
Many of the formulas for the inverse transform have been stated with-
out proof. This includes the Heaviside Expansion Theorem (6.16) and the
formula at a jump discontinuity (6.10). If you are interested in learning
about the more theoretical aspects of the subject, you might want to
consult Davies [2002] or, if you are more adventurous, Widder [1941].

Exercises
1. Use the Laplace transform to find the solution of the IVP.
a) 2y ′ + y = 1, y(0) = 2
b) 3y ′ = −y + e−t , y(0) = 21
c) y ′′ + y ′ − 2y = 0, y(0) = 0, y ′ (0) = −1
d) y ′′ − 6y ′ + 9y = 0, y(0) = 0, y ′ (0) = 2
e) 5y ′′ − y ′ = 0, y(0) = −1, y ′ (0) = −1
f) 4y ′′ + y = 0, y(0) = −1, y ′ (0) = −1
g) y ′′ − 2y ′ + 2y = 0, y(0) = −1, y ′ (0) = −1
h) y ′′ + 2y ′ + 5y = 0, y(0) = 0, y ′ (0) = −6
2. Use the Laplace transform to find the solution of the IVP.
a) y ′′ + y ′ − 2y = 12t, y(0) = 0, y ′ (0) = 0
b) y ′′ + 4y = 8t2 , y(0) = 0, y ′ (0) = 0
c) y ′′ − y ′ = 2 sin t, y(0) = 0, y ′ (0) = 0
d) y ′′ + 3y ′ = 3t + 1, y(0) = 0, y ′ (0) = 0
e) y ′′ − 2y ′ + 5y = 5 − 4e−t , y(0) = 0, y ′ (0) = 0
3. For the following, find the transfer function H(s) and then write down
the resulting particular solution. You do not need to evaluate the
integral.
a) y ′ + 3y = ln(1 + 3t)
√
b) y ′′ + 9y = 1 + t
c) 2y ′′ + 3y ′ − 2y = 1/(1 + t)
d) y ′′ + 2y ′ + 5y = sin(1 + t2 )
4. Proceeding as in Example 3, find the solution of the following IVPs.
a) y ′ + 3y = ln(1 + 3t), where y(0) = 1
√
b) y ′′ + 9y = 1 + t, where y(0) = 1 and y ′ (0) = 0
c) 2y ′′ + 3y ′ − 2y = 1/(1 + t), where y(0) = 2 and y ′ (0) = −3
d) y ′′ + 2y ′ + 5y = sin(1 + t2 ), where y(0) = 0 and y ′ (0) = 2
164 Chapter 6. Laplace Transform

6.5 Solving Equations with Non-Smooth Forcing

The next example considers how to solve a differential equation with a
discontinuous forcing function. This is a situation that is not uncommon
in applications.

Example: Solve y ′′ + 3y ′ + 2y = f (t), where y(0) = 1, y ′ (0) = −1, and

(
2 if 0 ≤ t ≤ 3,
f (t) =
0 if 3 < t.
Answer: Taking the Laplace transform of the differential equation,
L(y ′′ + 3y ′ + 2y) = L(f ). (6.27)
Using the linearity of the transform, and the derivative formulas
(6.18) and (6.19),
L(y ′′ + 3y ′ + 2y) = L(y ′′ ) + 3L(y ′ ) + 2L(y)
= s2 Y − y ′ (0) − sy(0) + 3 sY − y(0) + 2Y

= (s2 + 3s + 2)Y − s − 2.
R3
Also, L(f ) = 0 2e−st dt = 2 1−e−3s /s. Consequently, from (6.27),

we have that
2 
(s + 1)(s + 2)Y = s + 2 + 1 − e−3s ,
s
which means that
1 2 
−3s

Y = + 1−e . (6.28)
s + 1 s(s + 2)(s + 1)
To determine the inverse transform, using Property 8 from Table
6.1, L−1 (1/(s + 1)) = e−t . Also, from Property 11,
 1 
L−1 = e−t − e−2t .
(s + 2)(s + 1)
Consequently, using Properties 5 and 15 (respectively),
 2

L−1 1 − e−3s
s(s + 2)(s + 1)
−1
 2 
−1
 2 −3s

=L −L e
s(s + 2)(s + 1) s(s + 2)(s + 1)
Z t Z t−3
= 2 (e−r − e−2r )dr + 2H(t − 3) (e−r − e−2r )dr
0 0
= −2e−t + e−2t + 1 + H(t − 3)(1 − 2e3−t + e−2t+6 ).
The resulting solution is therefore
 
y = 1 + e−2t − e−t − 1 + e−2(t−3) − 2e−(t−3) H(t − 3). 
6.5. Solving Equations with Non-Smooth Forcing 165

A comment needs to be made about the mathematical correctness

of the solution we just derived. Namely, y(t) and y ′ (t) are defined and
continuous for 0 ≤ t < ∞, but y ′′ (t) is not continuous at t = 3 (it is,
however, continuous everywhere else). This throws into question whether
the differential equation y ′′ + 3y ′ + 2y = f (t) is defined at t = 3. The
way this needs to be interpreted is that the differential equation holds for
0 < t < 3, and then again for 3 < t < ∞. The discontinuity in the forcing
function effectively resets the problem at t = 3. One approach to dealing
with this is to break the problem into two IVPs, one for 0 < t < 3, and
another for 3 < t < ∞. By using the Laplace transform we have been
able to avoid having to do this. This is possible because the continuity
requirements to use (6.19) are satisfied for this problem.

6.5.1 Impulse Forcing

The idea underlying impulse forcing is that the force is fairly intense but it
occurs over a short time interval. Writing the interval as t0 −ε < t < t0 +ε,
we are considering the situation of when the forcing has the form

 0
 if 0 ≤ t ≤ t0 − ε,
f (t) = d(t) if t0 − ε < t < t0 + ε, (6.29)

0 if t0 + ε ≤ t.


With this, the solution of y ′ = f , where y(0) = 0, is



 R 0 if 0 ≤ t ≤ t0 − ε,
t
y= d(r)dr if t0 − ε < t < t0 + ε, (6.30)
 t0 −ε
D if t0 + ε ≤ t,


where Z t0 +ε
D= d(r)dr.
t0 −ε

We are assuming that the forcing interval is very short, but D is large
enough to be meaningful. To put this in physical terms, it is as if we are
hitting the system with a hammer.
There is a mathematical idealization for a concentrated force that
makes solving the problem easier than trying to use a formulation as in
(6.29). This is done by introducing what is known as the delta function.

Delta Function. The delta function δ(t) is defined to have the following
properties:

1. Given any t0 ,
δ(t − t0 ) = 0, when t 6= t0 . (6.31)
166 Chapter 6. Laplace Transform

2. Given any continuous function g(t), and assuming a < t0 < b:

Z b
δ(t − t0 )g(t)dt = g(t0 ), (6.32)
a

and
t0 b
1
Z Z
δ(t − t0 )g(t)dt = δ(t − t0 )g(t)dt = g(t0 ) . (6.33)
a t0 2

As an example of how the delta function is used, instead of using

(6.29), the forcing is assumed to have the form f (t) = Dδ(t − t0 ). This
means we are assuming that there is a delta forcing at t0 with strength D.
With this, the differential equation becomes

y ′ = Dδ(t − t0 ),

where y(0) = 0. The solution of this IVP is

Z t
y= Dδ(r − t0 )dr.
0

To evaluate this, first note that if 0 ≤ t < t0 , then from (6.31), y(t) = 0.
If t = t0 , then from (6.33), y(t0 ) = D/2. Lastly, when t0 < t, then from
(6.32), y(t) = D. Consequently, the solution is

 0
 if 0 ≤ t < t0 ,
1
y=
 2D if t = t0 , (6.34)
D if t0 < t.


Except for the very small time interval t0 − ε < t < t0 + ε, this solution
is the same as the one in (6.30). Moreover, the above solution is consis-
tent with what is obtained using the line integral formula for the inverse
Laplace transform.
The rationale for the stated properties of the delta function can be
explained by considering the case of when d is constant. The assumption is
that the total force D, what is known as the impulse, remains fixed as the
time interval decreases (see Figure 6.3). This requires that d = D/(2ε).
In other words, the magnitude of the force increases as the time interval
decreases. Consequently, in the limit, the forcing is zero if t 6= t0 and it is
infinite at t = t0 . This explains (6.31), and it also explains why you will
see the statement that δ(0) = ∞. This limit can also be used to explain
(6.32). Finally, it is being assumed that the impulse forcing is symmetric
about t0 , as it is in the case of when d is constant, and this gives us (6.33).
6.5. Solving Equations with Non-Smooth Forcing 167

Figure 6.3. A fixed impulse, applied over the time interval t0 − ε < t < t0 + ε,
used to explain the stated properties of the delta function.

Example 1: If f (t) = δ(t − a), where a > 0, find L(f ).

Answer: Using (6.32),
Z ∞
δ(t − a)e−st dt


L δ(t − a) =
0

= e−as .

This is Property 13 in Table 6.1. 

Example 2: If f (t) = 5δ(t − 1) − 9δ(t − 2) + δ(t − 3), find L(f ).

Answer: Using linearity and Property 13 in Table 6.1,




L(f ) = 5L δ(t − 1) − 9L δ(t − 2) + L δ(t − 3)

= 5e−s − 9e−2s + e−3s . 

Example 3: Solve y ′′ + y = 2δ(t − 15), where y(0) = 0 and y ′ (0) = 0.

Answer: Taking the Laplace transform of the differential equation
gives L(y ′′ +y) = L(2δ(t−15)). Using the linearity of the transform,
and the derivative formula (6.19), we get that

s2 Y − sy(0) − y ′ (0) + Y = 2e−15s .

From the given initial conditions, and solving for Y ,

2
Y = e−15s .
s2 +1
Since
2 
L−1 = 2 sin t,
s2 + 1
then using Property 5, from Table 6.1,

y = 2 sin(t − 15)H(t − 15). 

168 Chapter 6. Laplace Transform

Rt
Example 4: Evaluate −∞ δ(r)dr, for −∞ < t < ∞.
Answer: This can be answered by considering three cases. First, if
t < 0, then δ(r) = 0 for −∞ < r ≤ t, and we conclude the integral
is zero. If t > 0, then from (6.32) with g(t) = 1, the integral is equal
to one. Finally, when t = 0, from (6.33) we get the value of 1/2.
Therefore, we have that

Z t
H(t) = δ(r)dr.
−∞

In this sense we can write that

H ′ (t) = δ(t). 

Mathematical Tidbits

As you likely noticed, δ(t) is not actually a function. The more accurate
statement is that it is a distribution, or a generalized function. There are
various ways to obtain a mathematically rigorous definition of δ(t), using
limits or test functions. How limits are used was explained very briefly
earlier. This will not be pursued any further, but the question does arise
as to what is permitted when using the delta function. As demonstrated
in Example 2, linear combinations of delta functions are allowed. It is
also possible to both differentiate and integrate a delta function. What
should be avoided is using a discontinuous g(t) in (6.32), and Exercise 4 is
an example why. Also, what is not allowed, generally, involves nonlinear
operations. So, expressions such as δ(t − 1)δ(t − 2), δ(t − 1)/δ(t − 2), and
sin(δ(t−1)) are not allowed. If you are interested in the various properties
of the delta function, you might look at its Wikipedia page.
The nonstandard nature of the delta function amplifies a complication
with the Laplace transform at t = 0 that needs to be mentioned. It is
not uncommon in certain applications to use the forcing function f (t) =
δ(t), which means that it is located at t = 0. This puts it at the lower
point of integration for the Laplace transform.

The resulting integral can
be evaluated using (6.33), giving
L δ(t) = 1/2. However, it is almost
universally stated that L δ(t) = 1. One way to explain this involves
continuity, in the sense that this is what you obtain from Property 13,
in
+
Table 6.1, when letting a → 0 . On the other hand, one gets L δ(t) =
0
when letting a → 0− . This has lead those determined to obtain L δ(t) =
1 to find some rather creative ways to redefine the Laplace transform.
What this involves is not considered here, but if you are interested in
learning more about this issue, you should consult Hoskins [2009].
Exercises 169

Exercises
1. Use the Laplace transform to find the solution of the IVP.
a) y ′ + 4y = 3H(t − 1), y(0) = 1
b) 2y ′ − y = 1 − H(t − 4), y(0) = −1
c) y ′ + y = 2δ(t − 3), y(0) = −1
d) y ′ − 4y = 2H(t − 2) − δ(t − 1), y(0) = 0
e) y ′′ − y ′ − 6y = 3H(t − 5), y(0) = 0, y ′ (0) = 0
f) y ′′ + 4y = 3H(t − 4) − 3H(t − 2), y(0) = 0, y ′ (0) = 0
g) y ′′ − 4y ′ = 3δ(t − 1), y(0) = 0, y ′ (0) = 0
h) y ′′ + y = δ(t − 3) − 2δ(t − 2), y(0) = 0, y ′ (0) = 0
2. There are usually multiple ways to find an inverse transform, and this
exercise illustrates this by reconsidering (6.28).
a) Using partial fractions, the assumption is
2 A B C
= + + .
s(s + 2)(s + 1) s s+2 s+1
Find A, B, and C, and then determine the inverse transform.
b) Find the inverse transform of the function in part (a) but use Prop-
erty 12, with n = 3.
3. Show that the following identities hold for the delta function. Do
this by showing that when the left and right sides of the equation are
inserted into (6.31)-(6.33), that they produce the same result.
a) δ(a(t − t0 )) = a1 δ(t − t0 ), for a > 0
b) δ(t0 − t) = δ(t − t0 )
c) If g(t) is continuous, then g(t)δ(t − t0 ) = g(t0 )δ(t − t0 ) .
4. In quantum physics there are occasions when the coefficients of the
differential equation contain delta functions. The point of this exercise
is to demonstrate that care is needed in such situations.
a) Consider the problem of solving
y ′ (t) = δ(t − t0 )y(t), for t > 0,
where t0 > 0 and y(0) = 1. Using separation of variables, and
Example 4, find the solution. Make sure to determine its value for
0 ≤ t < t0 , for t = t0 , and for t0 < t. For the record, this is the
correct solution of this problem.
b) By simply integrating the differential equation in part (a), and then
using the initial condition, one gets that
Z t
y(t) = 1 + y(r)δ(r − t0 )dr.
0
170 Chapter 6. Laplace Transform

Not thinking too hard about the situation, and using (6.31)-(6.33),
explain how you might conclude that

 1
 if 0 ≤ t < t0 ,
y= 2 if t = t0 ,

 3 if t0 < t.

This differs from the solution for part (a). Where is the error made
in the derivation of the above solution?

6.6 Solving Linear Systems

The Laplace transform can be used to solve a linear system of differential
equations, and this is often the approach taken for what are known as
state space models in engineering. It is relatively easy to do this, and to
explain why, suppose we want to solve

x′ = ax + by + f (t), (6.35)
′
y = cx + dy + g(t), (6.36)

where x(0) = x0 and y(0) = y0 . Taking the Laplace transform of each

equation, and using (6.18), we get

sX − x0 = aX + bY + F, (6.37)
sY − y0 = cX + dY + G, (6.38)

where X, Y , F , and G are the Laplace transforms of x, y, f , and g, re-

spectively. What we have shown is that if the original differential equation
is written as x′ = Ax + f , then the transformed equation is

sX − x0 = AX + F, (6.39)

where X = L(x), F = L(f ), and x0 = x(0). This is the form obtained in

the general case of when there are n equations, and A is an n×n constant
matrix.
The next step is to solve for X and Y , and then attempt to find the
inverse transforms. How hard it is to find the inverse transforms depends
on f and g.

Example 1: Using the Laplace transform, solve

x′ = x − y,
y ′ = 4x − 2y,

where x(0) = 1 and y(0) = −1.

6.6. Solving Linear Systems 171

Answer: From (6.37) and (6.38) the transformed equations are

sX − 1 = X − Y,
sY + 1 = 4X − 2Y.

From the first equation, Y = 1 + (1 − s)X, and after substituting

this into the second equation, and simplifying, one finds that
s+3
X= .
s2 +s+2
√
Since s2 + s √+ 2 = (s − s1 )(s − s2 ), where s1 = (−1 + i 7)/2 and
s2 = (−1 − i 7)/2, then from Property 11 of Table 6.1,
 
−1 −t/2 5
x = L (X) = e cos(ωt) + sin(ωt) ,
2ω
√
where ω = 7/2. To find y we can either find the inverse transform
for Y , or we can use the first differential equation. The latter option
is easiest, and so

y = x − x′
 
−t/2 11
=e − cos(ωt) + sin(ωt) . 
2ω

Example 2: Using the Laplace transform, solve

x′ = 3x − 6y + f (t),
y ′ = x − 4y + g(t),

where x(0) = 0 and y(0) = 0. Also, f (t) and g(t) are continuous
functions.
Answer: From (6.37) and (6.38) the transformed equations are

sX = 3X − 6Y + F,
sY = X − 4Y + G.

From the second equation, X = (s + 4)Y − G, and after substituting

this into the first equation, and simplifying, one finds that
s−3 1
Y = G(s) + 2 F (s) .
s2 + s − 6 s +s−6
The convolution theorem is going to be used in finding the inverse
transform, and in preparation for this note that, using Property 11
of Table 6.1,
 1   1  1 
L−1 2 = L−1 = e2t − e−3t ,
s +s−6 (s − 2)(s + 3) 5
172 Chapter 6. Laplace Transform

and
s − 3  1  −3t

2t

L−1 = 6e − e .
s2 + s − 6 5
So, using the convolution theorem, which is Property 2 of Table 6.1,
 1  Z t 1 
−1 2(t−r) −3(t−r)
L F (s) = e − e f (r)dr,
s2 + s − 6 0 5

and
t
s−3 1  −3(t−r)
  Z 
L −1
G(s) = 6e − e2(t−r) g(r)dr.
s2 + s − 6 0 5
Therefore, the solution is
1 t 2(t−r)
Z
− e−3(t−r) f (r)dr


y(t) = e
5 0
1 t
Z
6e−3(t−r) − e2(t−r) g(r)dr.


+
5 0
To find x you can either find the inverse transform for X, or you
can use the second differential equation (similar to what was done
in the previous example). 

6.6.1 Chapter 4 versus Chapter 6

In Chapter 4 we solved problems as in Example 1 using an eigenvalue ap-
proach. In contrast, using the Laplace transform avoids this and it solves
the problem directly. Moreover, in Example 2, using the Laplace trans-
form the inhomogeneous problem was solved with little fanfare. This was
not done in Chapter 4 as it would have involved introducing either unde-
termined coefficients or variation of parameters. The apparent conclusion
is that linear systems are more easily solved using the Laplace transform
than the eigenvalue approach. This is true for systems with two equations
and the reason is that it is relatively easy to solve for X and Y . However,
for larger systems the advantage switches to the eigenvalue approach. The
reason is that larger systems are almost always solved numerically (i.e.,
using a computer). The eigenvalue approach provides a representation of
the solution (4.25), and it is relatively easy to compute the terms in that
expression. In contrast, from (6.39),
 −1
X = sI − A (F + x0 ).

This requires finding the inverse matrix and then trying to determine the
inverse transform of the resulting formula for X. There are ways this can
be done, such as using a geometric series expansion for (sI − A)−1 , but
how this is carried out is beyond the purview of this textbook. Those
interested might want to look at Friedland [2005] and Cohen [2007].
Exercises 173

Exercises
1. Use the Laplace transform to find the solution of the IVP, with
 
4
x(0) = .
−1
! ! !
′
−1 6 ′
3 1 ′
2 0
a) x = x c) x = x e) x = x
1 0 1 3 −1 2
! ! !
′
0 41 ′
2 1 ′
1 1
b) x = x d) x = x, f) x = x
1 0 6 3 −4 1

2. This exercise uses the Laplace transform to solve

x′ = ax + by,
y ′ = cx + dy,

where x(0) = x0 and y(0) = y0 .

a) Taking the Laplace of the above equations, and then solving for X
and Y , show that
1

X= (s − d)x0 + by0 ,
(s − r1 )(s − r2 )
1

Y = cx0 + (s − a)y0 .
(s − r1 )(s − r2 )

b) What values of a, b, c, and d result in r1 and r2 being real and r1 6=

r2 ? Assuming this is the case, use the inverse Laplace transform to
find x and y.
c) What values of a, b, c, and d result in r1 and r2 being complex
(with a nonzero imaginary part)? Assuming this is the case, use
the inverse Laplace transform to find x and y.
d) What values of a, b, c, and d result in r1 = r2 ? Assuming this is
the case, use the inverse Laplace transform to find x and y.
3. As defined in Section 6.4.1, the transfer function H(s) is obtained from
(6.39) by setting x0 = 0 and solving for X. The result is

H(s) = (sI − A)−1 .

Find the transfer function for the following systems.

174 Chapter 6. Laplace Transform

! !
−1 6 2 0
a) x′ = x c) x′ = x
1 0 −1 2
! !
1
0 4 1 −4
b) x′ = x d) x′ = x
1 0 1 1
Chapter 7

Partial Differential
Equations

A partial differential equation is simply a differential equation with

more than one independent variable. It is typical that the independent
variables are time (t) and space (x). If u(x, t) is the dependent variable,
then examples of partial differential equations (PDEs) are

Advection Equation: ut + aux = 0

Diffusion Equation: ut = Duxx
Wave Equation: utt = c2 uxx .

Each of these PDEs is linear and homogeneous. Also, the advection equa-
tion is first order, while the other two are second order.
Subscripts are used in the above PDEs to indicate partial differentia-
tion. There are two other ways this can be done that are very common.
First, there is the form used in calculus, and examples are

∂u ∂u ∂2u ∂2u ∂2u

, , , , .
∂t ∂x ∂t2 ∂t∂x ∂x2
A more contemporary notation is to abbreviate the above expressions,
and write
∂t u , ∂x u , ∂t2 u , ∂t ∂x u , ∂x2 u .
All three forms will be used in this chapter.
In this chapter, at the start, the method of separation of variables is
used to solve PDEs. Later, in Section 7.7, it will be shown how separation
of variables leads to another approach, called the Galerkin method. It is
also possible to use the Laplace transform to solve many of the problems
considered in this chapter, although that will not be pursued here.
Introduction to Differential Equations, M. H. Holmes, 2020

175
176 Chapter 7. Partial Differential Equations

7.1 Balance Laws

The PDEs listed above are the mathematical consequence of a balance
law, much like the ODEs obtained for simple harmonic motion in Section
3.10, and the various modeling examples in Section 2.3. For example,
the wave equation describes the vertical displacement u(x, t) of an elastic
string. The PDE is a force balance equation coming from Newton’s second
law F = ma. In this case, the acceleration is a = utt , and F is the vertical
component of the restoring force in the string due to its being stretched.
In contrast, the diffusion equation can be used to determine the den-
sity, or concentration, of objects moving along the x-axis due to Brownian
motion. The balance law in this case is the requirement that the total
number of objects is constant, which means that if one region experiences
an increase, then this is balanced by a decrease in other regions. In older
textbooks this equation is usually identified as the heat equation. How-
ever, it has far more applicability than heat propagation, and since about
1950 it is more often referred to as the diffusion equation in the research
literature.
Explaining the physical and mathematical assumptions underlying the
derivation of PDEs is outside the purview of this textbook. If you are
interested in this you should consult Holmes [2019].

7.2 Boundary Value Problems

The PDEs listed above involve the spatial variable x. Consequently, it
is worth first considering how to solve an ODE involving x. A typical
example is to find the function u(x) that satisfies

u′′ − 4u = 0, for 0 < x < 2, (7.1)

where
u(0) = 1, (7.2)
and
u(2) = −3. (7.3)
This is called a boundary value problem (BVP), and it consists of a
differential equation and two boundary conditions, one at each end
of the spatial interval. Because this involves a linear differential equation
with constant coefficients, the methods developed in Chapter 3 can be
used to solve it. So, assuming that u = erx , and then substituting this
into the differential equation (7.1), you obtain the characteristic equation
r2 = 4. The two solutions are r1 = −2 and r2 = 2, which means that the
general solution of (7.1) is

u = c1 e−2x + c2 e2x .
7.2. Boundary Value Problems 177

To satisfy the boundary condition at x = 0 we need c1 + c2 = 1 and

to satisfy the boundary condition at x = 2 we need c1 e−4 + c2 e4 = −3.
Solving these two equations yields

e4 + 3 e−4 + 3
c1 = and c2 = − .
e4 − e−4 e4 − e−4

The other methods derived in Chapter 3 are easily modified to solve

BVPs. As will be demonstrated in Example 1, the method of undeter-
mined coefficients can be used to solve a BVP. However, a complication
can arise as it is possible for the boundary conditions to be incompati-
ble with the differential equation. If this happens then the BVP has no
solution, and Example 2 is a demonstration of when this can happen.

Example 1: Solve u′′ − 3u′ + 2u = 4x, where u(0) = 3 and u(4) = 0.

Step 1: The associated homogeneous equation is u′′ − 3u′ + 2u = 0.

Assuming u = erx , one gets the characteristic equation r2 − 3r + 2 =
0. The roots are r = 1 and r = 2, and so uh = c1 ex + c2 e2x .
Step 2: To find a particular solution, we assume that u = Ax + B.
From the differential equation, we get that

2Ax − 3A + 2B = 4x.

Equating the respective coefficients, 2A = 4 and −3A + 2B = 0.

Solving these two equation yields A = 2 and B = 3.
Step 3: The general solution is

u = 2x + 3 + c1 ex + c2 e2x .

Step 4: For u(0) = 3 we need c1 + c2 = 0, and for u(4) = 0 we need

11 + c1 e4 + c2 e8 = 0. Consequently, c1 = −c2 = 11/(e4 (e4 − 1)), and
the resulting solution is

11 
x 2x

u = 2x + 3 + e − e . 
e4 (e4 − 1)

Example 2: Show that u′′ + u = 0, where u(0) = 1 and u(π) = −3, has
no solution.

Ans: Assuming u = erx gives r2 = −1, from which we get the

general solution u = c1 cos x + c2 sin x. To satisfy u(0) = 1 we need
c1 = 1 and to satisfy u(π) = −3 we need c1 = 3. This is not possible,
and so the BVP has no solution. 
178 Chapter 7. Partial Differential Equations

7.2.1 Eigenvalue Problems

We are going to have to consider a particular type of BVP when we solve
a PDE. An example is the problem of solving

u′′ − λu = 0, for 0 < x < 1, (7.4)

where
u(0) = 0, (7.5)
and
u(1) = 0. (7.6)
The function u = 0 is a solution, but what we want to know is whether
there are nonzero solutions. To be specific, is it possible to find values
of the constant λ so there are solutions that are not identically zero?
This is the same question asked when solving the eigenvalue problem
Aa = λa. In other words, finding u(x) and λ is an eigenvalue problem.
In this context, the u’s are called eigenfunctions, and the λ’s are the
eigenvalues. A distinctive difference from the matrix eigenvalue problem
is that there can be an infinite number of eigenvalues for an eigenvalue
BVP.
Finding λ and u is not hard. As usual, assuming that u = erx , then
the characteristic
√ equation coming from (7.4) is r2 − λ = 0. This means
that r = ± λ. Assuming λ is a real number then we have the following
three cases:
√ √
λ > 0 : In this case, the general solution is u = ae λx +be− λx
√ . To satisfy
√
u(0) = 0 we need a+b = 0, and for u(1) = 0√we need√ae λ +be− λ =
0.√ So, b = √
−a, and this means that a(e λ − e− λ ) = 0. Since
e λ 6= e− λ when λ > 0, the conclusion is that a = 0, and this
means we just get the zero solution.

λ = 0 : The general solution of (7.4) is u = a + bx. To satisfy u(0) = 0

we need a = 0, and for u(1) = 0 we need a + b = 0. So, a = b = 0
and this means we just get the zero solution.

λ < 0 : Setting λ = −k 2 , where k > 0, then r = ±ik. This means that

the general solution of (7.4) is

u(x) = a cos(kx) + b sin(kx).

To satisfy u(0) = 0 we need a = 0. To satisfy u(1) = 0 we need

b sin(k) = 0. To obtain a not identically zero solution for u(x) we
take k so that sin(k) = 0. This holds if any one of the following
values are used:
k = π, 2π, 3π, . . . .
Exercises 179

The conclusion is that the eigenfunctions are

un (x) = bn sin(nπx), (7.7)

where bn is an arbitrary nonzero constant, and the associated eigenvalues

are
λn = −(nπ)2 , (7.8)
for n = 1, 2, 3, . . ..

Skipping the Two Real Roots Case

An observation can be made that will simplify solving an eigenvalue prob-

lem. In the above example, when there were two real-valued solutions for
r, we ended up with the zero solution. This always happens. So, this
case will often be skipped and the stated reason will be that it corre-
sponds to two real roots. For example, if the characteristic equation is
r2 = −λ, then to skip the two real roots case it will be assumed that
λ ≥ 0. Similarly, 2
1
√ if the characteristic equation is r + λr + 4 = 0, then
2
r = 2 (−λ ± λ − 16), and so skipping the two real roots case means
that it is assumed that λ2 ≤ 16.

Rayleigh Quotient

It is possible to show that the eigenvalues for the above BVP must be
negative, without having to first derive the formula for them. This can
be done using what is called the Rayleigh quotient, and this is explained
in Exercise 4. In fact, the steps in this exercise can be modified to also
prove that the eigenvalues must be real-valued, which is an assumption
we made in solving the eigenvalue problem.
The Rayleigh quotient is more than a theoretical tool as it plays an im-
portant role when studying mechanical vibrations as well as when finding
quantum energy levels. It is also used extensively in scientific computing
when solving eigenvalue problems.

Exercises
1. Solve the given BVP.
a) u′′ − 4u = 0, for 0 < x < 2; u(0) = 0 and u(2) = 1.
b) u′′ + u = 0, for 0 < x < 1; u(0) = 0 and u(1) = −1.
c) u′′ + u′ + u = 0, for 0 < x < 1; u(0) = 0 and u(1) = 1.
d) u′′ − u = 5, for 0 < x < 2; u(0) = 0 and u(2) = 0.
e) u′′ + u′ = x, for 0 < x < 1; u(0) = 0 and u(1) = 0.
2. Show that the given BVP has no solution.
180 Chapter 7. Partial Differential Equations

a) u′′ + 9u = 0, for 0 < x < π; u(0) = 2 and u(π) = 1.

b) 4u′′ + u = 0, for 0 < x < π; u(0) = −1 and u′ (π) = 4.
c) 4u′′ + π 2 u = 0, for 0 < x < 1; u′ (0) = 0 and u(1) = −3.
3. Find the eigenvalues and eigenfunctions of the given BVP. You can use
the “skip the two real roots” simplification, just make sure to state the
resulting assumption on λ.
a) u ′′ = λu, for 0 < x < 1; u(0) = 0 and u ′ (1) = 0.
b) u ′′ = λu, for 0 < x < 4; u ′ (0) = 0 and u ′ (4) = 0.
c) u ′′ + λu ′ + u = 0, for 0 < x < 4; u(0) = 0 and u(4) = 0.
d) u ′′ + u ′ = λu, for 0 < x < 1; u(0) = 0 and u(1) = 0.
e) u ′′ + λu = 0, for 0 < x < 1; u(0) = u(1) and u ′ (0) = u ′ (1). These
are called periodic boundary conditions.
4. This exercise explores the usefulness of what is known as the Rayleigh
quotient for the eigenvalue problem (7.4)-(7.6).
a) If you multiply (7.4) by u, and then integrate over the interval, you
get Z 1
(uu′′ − λu2 )dx = 0.
0
From this show that
Z 1 Z 1
2
λ u dx = − (u′ )2 dx.
0 0

The Rayleigh quotient for this problem is obtained when you solve
the above equation for λ.
b) Use part (a) to explain why, given an eigenfunction u(x), that the
associated eigenvalue must be negative.
c) The fundamental eigenfunction corresponds to the case of n = 1 in
(7.7). Taking b1 = 1, sketch u1 (x) for 0 ≤ x ≤ 1. On the same axes,
also sketch w(x) = 4x(1 − x).
d) Part (c) shows that w(x) can be used as an approximation of u1 (x).
Use w in the Rayleigh quotient to obtain an approximation for λ1 .

7.3 Separation of Variables

The solution method will be introduced by using it to solve a problem
involving the diffusion equation. This requires a correctly formulated
problem, and the one considered is to find the function u(x, t) that satisfies
∂2u

∂u 0 < x < L,
D 2 = , for (7.9)
∂x ∂t 0 < t.
In this equation, the positive constant D is the called the diffusion coef-
ficient. To complete the formulation we will prescribe the values of u at
7.3. Separation of Variables 181

the two endpoints, where x = 0 and x = L, and at the beginning, when

t = 0. Specifically, for boundary conditions it is assumed that

u(0, t) = 0, (7.10)

and
u(L, t) = 0. (7.11)
For the initial condition, it is assumed that

u(x, 0) = g(x), for 0 < x < L, (7.12)

where g(x) is a given function.

When using the method of separation of variables, you first find all
possible nonzero solutions of the PDE that satisfy the boundary condi-
tions. It is important to note that u = 0 is a possible solution of (7.9)
that also satisfies (7.10) and (7.11). What we want are the nonzero ones.
The fact that u = 0 is a solution of the PDE, and the boundary
conditions, is required for the method of separation of variables to work.
The reason is that this will enable us to use the principle of superposition.
So, if the left boundary condition is changed to, say, u(0, t) = 1, or the
PDE is changed to, say, Duxx = ut + x, then separation of variables
will not work. What is necessary in these cases is to first transform the
problem into one where u = 0 is a solution of the PDE and boundary
conditions. How this is done is considered in Sections 7.6 and 7.7.

7.3.1 Separation of Variables Assumption

The assumption is simply that

u(x, t) = F (x)G(t). (7.13)

Substituting this into the PDE (7.9) gives DF ′′ (x)G(t) = F (x)G ′ (t).
Separating variables yields
F ′′ (x) G ′ (t)
D = . (7.14)
F (x) G(t)
Now comes the key observation. The only way a function of x can equal
a function of t, since x and t are independent, is that the function of x is
a constant, the function of t is a constant, and the constants are equal.
In other words, there is a constant λ so that
F ′′ (x)
D = λ,
F (x)
and
G ′ (t)
= λ.
G(t)
182 Chapter 7. Partial Differential Equations

These can be rewritten as

DF ′′ (x) = λF (x) . (7.15)

and
G ′ (t) = λG(t) . (7.16)
The λ appearing here is called, not surprisingly, the separation con-
stant.

7.3.2 Finding F (x) and λ

The separation of variables assumption must be applied to the boundary
conditions. So, to have u(0, t) = 0, we need F (0)G(t) = 0. For this to
happen, and u not be identically zero, we require that F (0) = 0. Similarly,
we need F (L) = 0. Consequently, all-together, the function F (x) must
satisfy
DF ′′ (x) = λF (x) , for 0 < x < L, (7.17)
where
F (0) = 0 and F (L) = 0, (7.18)
The solution of this BVP depends on whether λ is zero or not. So, we
have two cases to consider.

λ = 0 : In this case (7.17) is F ′′ = 0, and so F (x) = a + bx. To satisfy

F (0) = 0 we need a = 0, and for F (L) = 0 we need b = 0. So, we
just get the zero solution in this case.

λ 6= 0 : Assuming F (x) = erx , then (7.17) reduces to Dr2 = λ. We will

skip the two real roots case, which means we only consider λ < 0.
Setting λ√= −k 2 , where k > 0, then Dr2 = −k 2 . This means that
r = ±ik/ D. The resulting general solution of (7.17) is
√ √
F (x) = a cos(kx/ D) + b sin(kx/ D).

To satisfy√F (0) = 0 we need a = 0. To satisfy F (L) = 0 we need

b sin(kL/ D) = 0. To obtain a function
√ F (x) that is not identically
zero, we take k so that sin(kL/ D) = 0. This holds if any one of
the following values are used:
√
kL/ D = π, 2π, 3π, . . . ,

or equivalently
√ √ √
π D 3π D 3π D
k= , , ,.... (7.19)
L L L
7.3. Separation of Variables 183

The conclusion is that the not identically zero solutions of (7.17) and
(7.18) are  nπx 
Fn (x) = bn sin , (7.20)
L
and  nπ 2
λn = −D , (7.21)
L
for n = 1, 2, 3, . . .. Also, bn is an arbitrary constant.

7.3.3 Finding G(t)

For λ = λn , the general solution of (7.16) is

Gn (t) = an eλn t , (7.22)

where an is an arbitrary constant. The function Gn (t) is not required to

satisfy the initial condition (7.12); that condition will be satisfied once we
determine the general solution.

7.3.4 The General Solution

We have shown that for any given n, the function un (x, t) = Fn (x)Gn (t)
is a solution of the PDE that satisfies the boundary conditions. Because
the PDE and boundary conditions are homogeneous, and the problem is
linear, the principle of superposition can be used (see page 5). There-
fore, the resulting general solution, that satisfies the PDE and boundary
conditions, is
X∞
u(x, t) = un (x, t),
n=1
or equivalently
∞
X  nπx 
u(x, t) = bn eλn t sin , (7.23)
L
n=1

where bn is an arbitrary constant, and λn is given in (7.21). In writing

this down, the constant an in (7.22) has been absorbed into the bn .

7.3.5 Satisfying the Initial Condition

It remains to satisfy the initial condition, which is u(x, 0) = g(x). Ac-
cording to our solution in (7.23), we need
∞
X  nπx 
bn sin = g(x). (7.24)
L
n=1

This is the equation that is used to determine the bn ’s. However, the
left-hand-side is an example of what is known as a Fourier series. More
184 Chapter 7. Partial Differential Equations

specifically, it is an example of a Fourier sine series. There are some

significant mathematical questions that arise here, one of which is whether
the series converges. This, and some related questions, are addressed in
the next section. For the moment, we simply state the conclusion. If g(x)
is continuous, except perhaps for a few jump discontinuities, then

2 L  nπx 
Z
bn = g(x) sin dx. (7.25)
L 0 L

7.3.6 Examples
Example 1: Suppose that D = 1, L = 2, and g(x) = 3 sin(πx). In this
case, from (7.21), λn = −(nπ/2)2 , and the resulting general solution
(7.23) is
∞  nπx 
2 2
X
u(x, t) = bn e−n π t/4 sin .
2
n=1

To satisfy the initial condition, it helps to notice that g(x) is one

of the sine functions in the series. To make this more evident, the

3
t=0
t = 0.05
Solution

t = 0.15
0

-3
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x-axis

Figure 7.1. Solution of the diffusion equation in Example 1. Shown is the

solution surface as well as the solution profiles at specific time values.
7.3. Separation of Variables 185

requirement that u(x, 0) = g(x) means that we need

∞
X  nπx 
bn sin = 3 sin(πx),
2
n=1

or equivalently
 πx   3πx 
b1 sin +b2 sin(πx)+b3 sin +b4 sin(2πx)+· · · = 3 sin(πx).
2 2
To satisfy this equation, take b2 = 3 and set all the other bn ’s to
zero. Therefore, the solution is
2
u(x, t) = 3e−π t sin(πx). (7.26)

This solution is shown in Figure 7.1, both as time slices and as the
solution surface for 0 ≤ t ≤ 0.24. 

Example 2: Suppose that in the previous example,

 πx   3πx 
g(x) = 3 sin − 4 sin + 5 sin(2πx).
2 2
This is an example of when g(x) involves the sum of three of the
sine functions in the series. The requirement is that
∞
X  nπx   πx   3πx 
bn sin = 3 sin − 4 sin + 5 sin(2πx).
2 2 2
n=1

To satisfy this we take b1 = 3, b3 = −4, b4 = 5, and all the other

bn ’s are zero. The resulting solution is
2
 πx  2
 3πx 
u(x, t) = 3e−π t/4 sin − 4e−9π t/4 sin
2 2
−4π 2 t
+ 5e sin(2πx). 

Example 3: Suppose that D = 1, L = 1, and

(
1 if 31 ≤ x ≤ 2
3 ,
g(x) = (7.27)
0 otherwise.

In this case, it is necessary to use (7.25) to find the bn ’s. Carrying

out the integration
Z 2/3
bn = 2 sin(nπx)dx
1/3
2  
= cos(nπ/3) − cos(2nπ/3) .
nπ
186 Chapter 7. Partial Differential Equations

1 t=0
t = 0.002
Solution

t = 0.03
0.5 t = 0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x-axis

Figure 7.2. Solution of the diffusion equation in Example 3. Shown is the

solution surface as well as the solution profiles at specific time values.

As for the solution, since λn = −(nπ)2 , then

∞
2 π2 t
X
u(x, t) = bn e−n sin(nπx). (7.28)
n=1

This solution is shown in Figure 7.2 for 0 ≤ t ≤ 0.1. 

Exercises
1. You are to find the solution of the diffusion problem for the following
initial conditions. Assume that L = 1 and D = 3. Note that you
should be able to answer this question without using integration.
a) g(x) = −4 sin(5πx).
b) g(x) = 6 sin(11πx).
c) g(x) = sin(πx) + 8 sin(4πx) − 10 sin(7πx).
d) g(x) = − sin(3πx) + 7 sin(8πx) + 2 sin(15πx).
e) g(x) = 4 sin(2πx) cos(πx).
Exercises 187

2. You are to find the solution of the diffusion problem for the following
initial conditions. Assume that L = 2 and D = 4.

a) g(x) = 1 −1 if 0 ≤ x ≤ 1
d) g(x) =
0 otherwise
b) g(x) = 2 + x
1 if 0 ≤ x ≤ 13

c) g(x) = cos(πx) e) g(x) =
2 otherwise

3. Find the solution of

∂2u

∂u 0 < x < 3,
9 2 = , for
∂x ∂t 0 < t,

where u(0, t) = 0, u(3, t) = 0, and u(x, 0) = −5x.

4. Find the solution of

2 0 < x < 2,
10 ∂x u = ∂t u , for
0 < t,

if 12 ≤ x ≤ 3

1 2
where u(0, t) = 0, u(2, t) = 0, and u(x, 0) =
0 otherwise.
5. Find the general solution of the following.
a) uxx = ut , for 0 < x < 1, with the boundary conditions u(0, t) = 0
and ux (1, t) = 0.
b) 4uxx = ut , for 0 < x < 1, with the boundary conditions ux (0, t) = 0
and u(1, t) = 0.
c) (1 + t)∂x2 u = ∂t u, for 0 < x < 1, with the boundary conditions
u(0, t) = 0 and u(1, t) = 0.
d) uxx = ut + e−t u, for 0 < x < 1, with the boundary conditions
u(0, t) = 0 and u(1, t) = 0.
6. Find the solution of the problem for the given initial condition.
a) Exercise 5(a), with u(x, 0) = 3 sin πx 9πx



2 −
7 sin 2 .
b) Exercise 5(b), with u(x, 0) = −5 cos 3πx 11πx


2 − 2 cos 2 .
c) Exercise 5(c), with u(x, 0) = 14 sin(10πx) + 30 sin(18πx).
d) Exercise 5(d), with u(x, 0) = −24 sin(3πx) − 12 sin(15πx).
7. Find the resulting ODEs obtained using separation of variables on the
given PDE.
a) (1 + x)uxx + tu = 7ut , assuming u = F (x)G(t)
b) r2 urr + rur + uθθ = 0, assuming u = R(r)Θ(θ)
c) ∂x (ex ∂x u) = (1 + x2 )∂t u, assuming u = F (x)G(t)
d) uzz + 3zuz = uyy + 9u, assuming u = Z(z)Y (y)
e) u2x + u2t = e−t u2 , assuming u = F (x)G(t)
188 Chapter 7. Partial Differential Equations

7.4 Sine and Cosine Series

To satisfy the initial condition for the diffusion problem considered in the
previous section, we were required to find the bn ’s so that
X∞  nπx 
g(x) = bn sin , for 0 < x < L. (7.29)
L
n=1
This is an example of a Fourier sine series. Finding the bn ’s is not hard.
However, this requires knowing what restrictions must be placed on g(x),
and so, this is where we begin.
One of the requirements is that g(x) is piecewise continuous for 0 ≤
x ≤ L. This means that g(x) is continuous on the interval except, possibly,
for a finite number of jump discontinuities. What a jump discontinuity
means is that g(x) is not continuous at the point, but the limits of g(x)
from the left, g(x− ), and from the right, g(x+ ), are defined and finite.
This is the requirement when 0 < x < L. For x = 0, then g(0+ ) must
be defined and finite, but it is not required to equal g(0). Similarly, for
x = L, g(L− ) must be defined and finite, but it is not required to equal
g(L). An example of a function with two jumps is given in (7.27). Also,
all of the functions in Exercise 2 in the previous section are piecewise
continuous.
A consequence of the assumption that g(x) is piecewise continuous is
that the integral in (7.25) is well-defined.

7.4.1 Finding the bn ’s

The working hypothesis is that the sine series converges, and we can
integrate it term-by-term. The reason for this assumption is that the
key for finding the coefficients is the integration formula: if m and n are
positive integers, then

Z L 
nπx   mπx   L

if m = n,
sin sin dx = 2 (7.30)
0 L L 
 0 if m 6= n.
The derivation of this formula is often done in calculus, and it involves
using the identity sin ax sin bx = 21 cos(a − b)x − cos(a + b)x .


To illustrate how (7.30) is used, suppose we want to find the value for,
say, b7 . Multiplying (7.29) by sin(7πx/L), and then integrating yields
Z L  7πx  ∞ Z L 
X 7πx   nπx 
g(x) sin dx = bn sin sin dx.
0 L 0 L L
n=1
According to (7.30), all of the integrals on the right are zero except when
n = 7. Consequently,
Z L  7πx  L
g(x) sin dx = b7 ,
0 L 2
7.4. Sine and Cosine Series 189

or equivalently
L
2  7πx 
Z
b7 = g(x) sin dx.
L 0 L
A similar result is obtained for the other bn ’s, and the resulting formula
is
2 L  nπx 
Z
bn = g(x) sin dx. (7.31)
L 0 L

7.4.2 Convergence Theorem

Proving a sine series converges requires more than just using the ratio
test, which is the way you prove a power series converges. The proof is
beyond the scope of this textbook, but the result is important for our
using a sine series when solving PDEs.

Sine Series Convergence Theorem. Assume that g(x) and g ′ (x) are
piecewise continuous for 0 ≤ x ≤ L, and the bn ’s are given in (7.31).
For 0 < x < L: If g(x) is continuous at x, then
∞
X  nπx 
g(x) = bn sin , (7.32)
L
n=1

and if g(x) has a jump discontinuity at x, then

∞
1 +  X  nπx 
g(x ) + g(x− ) = bn sin . (7.33)
2 L
n=1

At x = 0 or x = L: The sine series is zero when x = 0 or x = L.

In words, the theorem states that the sine series equals the function g(x)
at points in the interval where g(x) is continuous, and it equals the average
in the jump of g(x) at a jump discontinuity. At the endpoints, no matter
what the value of g(0) or g(L), the series sums to zero.

7.4.3 Examples
Finding a sine series is rather uneventful as it is simply a matter of eval-
uating the given formulas. The only concern is how hard it is to evaluate
the integrals to find the coefficients. So, in the examples below, a more
practical question is also considered. Namely, how many terms of the
series do you have to add together to obtain an accurate approximation
of the function g(x)? As will be seen, the answer depends on whether
the function is continuous, and whether it has the right values at the
endpoints.
190 Chapter 7. Partial Differential Equations

g(x)
1

0
0 1 2 3
x-axis

1
g (x)

0.5

0
0 1 2 3
x-axis
2
Sine Series

0
0 1 2 3
x-axis

Figure 7.3. The functions g(x), g ′ (x), and the function that the sine series
sums to for Example 1.

Example 1: Taking L = 3, suppose

(
x if 0 ≤ x ≤ 2 ,
g(x) =
1 if 2 < x ≤ 3.

Sketch g(x) and g ′ (x) for 0 ≤ x ≤ 3, and use this to explain why
they are piecewise continuous. Also, sketch the function that the
sine series for g(x) converges to for 0 ≤ x ≤ 3.
Answer: The functions g(x) and g ′ (x) are shown in Figure 7.3.
Although g(x) is not continuous at x = 2, and g ′ (x) is not defined
there, the left and right limits of both functions are defined and finite
at that point. Therefore, both functions are piecewise continuous
for 0 ≤ x ≤ 3. As for the sine series, for 0 < x < 3, it equals
g(x) except at the discontinuity, where it sums to the average in the
jump. So, at x = 2, it converges to 21 [g(2+ ) + g(2− )] = 21 (1 + 2) = 23 .
Finally, at x = 0, and at x = 3, the series sums to zero. The sketch
of the resulting function is given in Figure 7.3. 

√
Example 2: Taking L = 5, suppose g(x) = x. Are g(x) and g ′ (x)
piecewise continuous for 0 ≤ x ≤ 5?
Answer: The function g(x) is continuous for 0 ≤ x ≤ 5 (and it is,
7.4. Sine and Cosine Series 191

√
therefore, piecewise continuous). Its derivative g ′ (x) = 1/(2 x) is
continuous for 0 < x ≤ 5, but g ′ (0+ ) = ∞. Because this limit is not
finite, g ′ (x) is not piecewise continuous for 0 ≤ x ≤ 5. 

Example 3: For 0 ≤ x ≤ 1, find the sine series of

(
3x if 0 ≤ x ≤ 13 ,
g(x) = 3
2 (1 − x) if 31 < x ≤ 1.

Answer: Using (7.31), and integrating by parts,

Z 1/3 Z 1
bn = 6 x sin(nπx)dx + 3 (1 − x) sin(nπx)dx
0 1/3
9
= sin(nπ/3).
π 2 n2

Because g(x) is continuous, and g(0) = g(1) = 0, we have that

∞
X 9
g(x) = sin(nπ/3) sin(nπx), for 0 ≤ x ≤ 1. (7.34)
π 2 n2
n=1

1 1
g(x)

g(x)

0.5 0.5

N=1 N=3
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x-axis x-axis

1 1
g(x)

g(x)

0.5 0.5

N=9 N = 27
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x-axis x-axis

Figure 7.4. Comparison between the function g(x) in Example 3, shown with
the dashed blue curve, and the sine series approximation in (7.35), shown using a solid
red curve.
192 Chapter 7. Partial Differential Equations

When computing the value of the sine series it is necessary to pick

an N , and then use the approximation
N
X 9
g(x) ≈ sin(nπ/3) sin(nπx). (7.35)
2π 2 n2
n=1

The accuracy of this is shown in Figure 7.4. It is evident that for

smaller values of N the approximation is not very good, but it is
not bad for N = 27. 

Example 4: For 0 ≤ x ≤ 1, find the sine series of

(
1
1 if 0 ≤ x ≤ 4 ,
g(x) =
0 otherwise.

Also, sketch the function the sine series converges to for 0 ≤ x ≤ 1.

Answer: Using (7.31),
Z 1/4
2  
bn = 2 sin(nπx)dx = 1 − cos(nπ/4) .
0 nπ
From this we have that, except for x = 0 and x = 1/4,
∞
X 2  
g(x) = 1 − cos(nπ/4) sin(nπx). (7.36)
nπ
n=1

At x = 0 the series is zero, and at x = 1/4 the series sums to 1/2,

which is the average in the jump of g(x) at this point. The resulting
function is shown in Figure 7.5.
The resulting approximation is, given N ,
N
X 2  
g(x) ≈ 1 − cos(nπ/4) sin(nπx). (7.37)
nπ
n=1

1
Sine Series

0.5

0
0 0.25 0.5 0.75 1
x-axis

Figure 7.5. The function the sine series in Example 4 converges to for 0 ≤ x ≤ 1.
7.4. Sine and Cosine Series 193

1 N=4 1 N = 20
g(x)

g(x)
0.5 0.5

0 0

0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

x-axis x-axis

1 N = 40 1 N = 200
g(x)

g(x)
0.5 0.5

0 0

0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

x-axis x-axis

Figure 7.6. Comparison between the function g(x) for Example 4, shown
with the blue curve, and the sine series approximation in (7.37), shown using the red
curve.

The accuracy of this is shown in Figure 7.6. Because of the jump in

the function, the sine series requires a larger value of N than needed
in Example 3 to provide an accurate approximation. However, even
with a larger N , the series has difficulty in the immediate vicinity of
the jump. It has the same problem near x = 0 since the series is zero
at x = 0 but g(0) = 1. The larger oscillations near the jump points
are associated with what is called Gibbs phenomenon. As can
be seen in the figure, the region where these oscillations occur can
be reduced by taking larger values of N . However, the maximum
overshoot and undershoot on either side of the jump do not go to
zero. Instead, for 0 < x < L, they approach a value that is equal to
about 9% of the jump in the function. Because jump discontinuities
arise so often in applications, there has been considerable research
into how to remove the over and under shoots in the Fourier series
solution. One of the more well known methods involves filtering
them out, and an example is Fejér summation. More about this can
be found in Jerri [1998]. 

7.4.4 Cosine Series

Using separation of variables, it is not uncommon to end up with a cosine
series rather than a sine series. In this case, the initial condition requires
194 Chapter 7. Partial Differential Equations

finding the an ’s that satisfy

∞
1 X  nπx 
g(x) = a0 + an cos , for 0 < x < L. (7.38)
2 L
n=1

The convergence theorem for this is very similar to the one for the sine
series. First, the needed integration formula is, if m and n are integers,

L if m = n = 0,



Z L  nπx   mπx  
 L
cos cos dx = if m = n 6= 0, (7.39)
0 L L 
 2


0 if m 6= n.

The derivation of this formula is a straightforward
calculation using the

identity cos ax cos bx = cos(a − b)x + cos(a + b)x /2. This formula is
used in the same way the one for the sine series was used. Namely, if you
want to determine, say, a4 , you multiply (7.38) by cos(4πx/L) and then
integrate over the interval 0 ≤ x ≤ L. The resulting formula, for general
n, is
2 L  nπx 
Z
an = g(x) cos dx. (7.40)
L 0 L
This brings us to the next result.

Cosine Series Convergence Theorem. Assume that g(x) and g ′ (x)

are piecewise continuous for 0 ≤ x ≤ L, and the an ’s are given in (7.40).
If g(x) is continuous at x, then
∞
1 X  nπx 
g(x) = a0 + an cos . (7.41)
2 L
n=1

If g(x) has a jump discontinuity at x, and 0 < x < L, then

∞
1 +  1 X  nπx 
g(x ) + g(x− ) = a0 + an cos . (7.42)
2 2 L
n=1

At x = 0, the series sums to g(0+ ), and at x = L, the series sums to

g(L− ).

In words, the theorem states that the cosine series equals the function
g(x) at points in the interval where g(x) is continuous, and it equals the
average in the jump of g(x) at a jump discontinuity. At the endpoints, it
sums to the respective limit of g(x) at the endpoint.
7.4. Sine and Cosine Series 195

Example 5: For 0 ≤ x ≤ 1, find the cosine series of

(
x+1 if 0 ≤ x ≤ 12 ,
g(x) =
2 if 21 < x ≤ 1.

Answer: Using (7.40), if n 6= 0,

Z 1
an = 2 g(x) cos(nπx)dx
0
Z 1/2 Z 1
=2 (x + 1) cos(nπx)dx + 2 2 cos(nπx)dx
0 1/2
2  nπ
 1 nπ

= 2 2
cos − 1 − sin ,
n π 2 nπ 2
and when n = 0, a0 = 13/4. From this we have that, except for
x = 1/2,
∞
13 X h 2  nπ
 1 nπ
i
g(x) = + cos − 1 − sin cos(nπx).
8 n2 π 2 2 nπ 2
n=1
(7.43)
At x = 1/2 the series sums to the average in the jump in g(x), and
so it equals 7/4. The resulting function is shown in Figure 7.7.
The resulting approximation is, given N ,
N
13 X h 2  nπ
 1 nπ
i
g(x) ≈ + cos − 1 − sin cos(nπx).
8 n2 π 2 2 nπ 2
n=1
(7.44)
The accuracy of this is shown in Figure 7.8. As happened with
the sine series, in the immediate vicinity of the jump the series
oscillates. However, unlike Example 4, there are no oscillations at
the endpoints. 

2
Cosine Series

1.5

1
0 0.5 1
x-axis

Figure 7.7. The function the cosine series in Example 5 converges to for 0 ≤ x ≤ 1.
196 Chapter 7. Partial Differential Equations

2 2
g(x)

g(x)
1.5 1.5

N=4 N = 20
1 1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x-axis x-axis

2 2
g(x)

g(x)
1.5 1.5

N = 40 N = 200
1 1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x-axis x-axis

Figure 7.8. Comparison between the function g(x) for Example 5, shown
with the blue curve, and the cosine series approximation in (7.44), shown using the red
curve.

7.4.5 Differentiability
In using a sine or cosine series when solving a PDE, it is implicitly assumed
you can differentiate the series term-by-term. What this means is that it
is assumed that
∞ ∞
d X X d
pn (x) = pn (x).
dx dx
n=1 n=1

With this in mind, in Example 4, if you try this with (7.36), you get
∞
X
g ′ (x) =


2 1 − cos(nπ/4) cos(nπx). (7.45)
n=1
P
As you should recall, if an infinite series an converges, then it must be
true that an → 0 as n → ∞. The above series for g ′ (x) does not satisfy
this condition, and therefore it does not converge. In other words, you
can not differentiate (7.36) term-by-term. In contrast, for Example 3 you
can differentiate the series term-by-term. The theorem that explains this
states that if g(x) is continuous, and g ′ (x) is piecewise continuous, for
0 ≤ x ≤ L, then you can differentiate the cosine series term-by-term, but
to do this for a sine series you need an additional assumption [Tolstov and
Silverman, 1976]. An easy to use version of the needed assumption is that
g(0) = g(L) = 0. This holds for Example 3, and that is why term-by-term
7.4. Sine and Cosine Series 197

differentiation can be done with that sine series. For both the cosine and
sine series, if g(x) is not continuous, then term-by-term differentiation is
not possible without additional assumptions. Those interested in pursuing
this issue a bit further should look at Exercise 11.
The situation for term-by-term integration is better. Specifically, if
g(x) satisfies the requirements of the convergence theorem, its sine, and
cosine, series can be integrated term-by-term.
The next question is whether the potential non-differentiability of a
sine series means that we can not use it to solve the diffusion equation.
To explain why this is not a problem, consider the solution (7.43), which
is
∞
2 π2 t
X
u(x, t) = bn e−n sin(nπx).
n=1

As long as t > 0, the coefficients of this series are exponentially decreas-

ing functions of n2 . This, along with the fact that the series for g(x)
converges, guarantee that you can differentiate the series term-by-term
without reservation, as long as t > 0.

7.4.6 Infinite Dimensional

For vectors in R3 , the dot product is used to determine orthogonality. As

you should recall, x and y are orthogonal if x · y = 0. Also, any vector
in R3 can be written in terms of the three coordinate vectors i, j, and k.
This means that it is possible to write x as a linear combination of these
vectors: x = xi + yj + zk. In this sense, i, j, and k are a basis for R3 .
In fact, since i, j, and k are orthogonal to each other, they form what is
called an orthogonal basis. Because there are three vectors in the basis,
R3 is three dimensional.
Now, the Sine Series Convergence Theorem states when a function
g(x) can be written as a linear combination of the sine functions sin(πx/L),
sin(2πx/L), sin(3πx/L), · · · . There is also a dot product, or what is usu-
ally called an inner product, for the sine functions, and it involves the
integral appearing in the integration formula in (7.30). According to this
integration rule, the sine functions are orthogonal to each other. This
means that sin(πx/L), sin(2πx/L), sin(3πx/L), · · · is an orthogonal ba-
sis. Because this basis contains an infinite number of elements, the space
we are considering is infinite dimensional. This viewpoint gives rise to
what is called a Hilbert space, and these play a fundamental role in many
areas in science and engineering. For an introduction to Hilbert spaces
and partial differential equations, you might consult Gustafson [1999].
198 Chapter 7. Partial Differential Equations

Exercises
1. Sketch the graph of f (x) for 0 ≤ x ≤ 1. Also, determine whether f (x)
is continuous, piecewise continuous, or neither for 0 ≤ x ≤ 1.
 
1
 1 if 0 ≤ x ≤ 2
  0 if x = 0

1 3
a) f (x) = 2x if 2 < x ≤ 4 c) f (x) = ln x if 0 < x < 1
 3
 3
if 4 < x ≤ 1

 1 if x = 1
2
( (
1 if x = 14 , 21 , 43 , 1 0 if 0 ≤ x ≤ 12
b) f (x) = d) f (x) = 1 1
2 otherwise 2x−1 if 2 < x ≤ 1

2. Assuming that L = 2, explain why g(x) does not satisfy the conditions
stated in the Sine Series Convergence Theorem.

1
a) g(x) = x1/3 b) g(x) = tan x c) g(x) = x2 +4x−1

3. In the following, g(x) and g ′ (x) are piecewise continuous. Assuming

that L = 2, sketch the function to which the sine series converges, for
0 ≤ x ≤ 2.
(
a) g(x) = x 3 if x = 0, 21 , 1, 2
f) g(x) =
b) g(x) = ex x otherwise
c) g(x) = cos(πx)
(
−1 if x = 0, 31 , 2
( g) g(x) =
1 if 0 ≤ x ≤ 12 ex otherwise
d) g(x) =
−3 if 12 < x ≤ 2 
( 
 1 if 0 ≤ x ≤ 31
2 − x if 0 ≤ x ≤ 1 h) g(x) = 0 if 13 < x ≤ 34
e) g(x) =
0 if 1 < x ≤ 2 3 if 43 < x ≤ 2



4. Find the sine series for the functions in Exercise 3.

5. For the functions in Exercise 3, sketch the function to which the cosine
series converges, for 0 ≤ x ≤ 2.
6. Find the cosine series for the functions in Exercise 3.
7. For any given x from the interval 0 ≤ x ≤ 1, use the comparison test
to show that the series in (7.34) converges absolutely.
8. In this exercise let g(x) = x2 , for 0 ≤ x ≤ 1.
a) Find the cosine series for g(x).
b) For any given x from the interval 0 ≤ x ≤ 1, use the comparison
test to show that the series in part (a) converges absolutely.
7.5. Wave Equation 199

c) Using your result from part (a), show that

∞
X (−1)n+1 π2
= .
n2 12
n=1

9. In this exercise let g(x) = x, for 0 ≤ x ≤ 1.

a) Find the sine series for g(x).
b) Using your result from part (a), show that
π 1 1 1
= 1 − + − + ··· .
4 3 5 7
10. Find a function g(x) that is continuous for 0 ≤ x ≤ 1, except for a
jump discontinuity at x = 1/2, and which equals its sine series for
0 ≤ x ≤ 1.
11. This exercise deals with the restriction on term-by-term differentiabil-
ity of the sine series. This requires you to have read Section 6.5.1. If
the observation made in this exercise interests you, you might want to
look at Stakgold [2000].
a) Write the function g(x) in Example 4 in terms of the Heaviside
function H(x).
b) Using Example 4, from Section 6.5.1, what is g ′ (x)?
c) Using your result from part (b), what is the sine series for g ′ (x)?
How does this differ from the result in (7.45)?

7.5 Wave Equation

The problem involves finding the function u(x, t) that satisfies
2 ∂2u

2∂ u 0 < x < L,
c 2
= 2 , for (7.46)
∂x ∂t 0 < t,

where c is a positive constant. This PDE is known as the wave equation.

It applies, for example, to the vertical displacement u(x, t) of an elastic
string. This provides an interesting interpretation of the terms in the sine
series solution, and this is discussed in Section 7.5.2.
To complete the problem, the boundary conditions are

u(0, t) = 0, (7.47)

and
u(L, t) = 0. (7.48)
For the initial conditions, it is assumed that

u(x, 0) = g(x), for 0 < x < L, (7.49)

200 Chapter 7. Partial Differential Equations

and
ut (x, 0) = h(x), for 0 < x < L, (7.50)
where g(x) and h(x) are given functions. To avoid the complication with
differentiability, as described in Section 7.4.5, it is assumed that g(x)
and h(x) are smooth functions that satisfy the boundary conditions, and
g ′′ (0) = g ′′ (L) = 0.
As with the diffusion problem, separation of variables will be used to
find the general solution of the PDE and boundary conditions. After that,
the initial conditions will be satisfied. Also, you should notice, as with the
diffusion problem, the PDE and boundary conditions are homogeneous.
This is required for separation of variables to work.

Separation of Variables Assumption

Assuming
u(x, t) = F (x)G(t), (7.51)
and then substituting this into the PDE gives us

F ′′ (x) G ′′ (t)
c2 = . (7.52)
F (x) G(t)

Since the left-hand-side is only a function of x, and the right-hand-side is

only a function of t, we can conclude that there is a constant λ so that

c2 F ′′ (x) = λF (x) , (7.53)

and
G ′′ (t) = λG(t) . (7.54)

Finding F (x) and λ

The separation of variables assumption must be used on the boundary

conditions. So, to have u(0, t) = 0, we need F (0)G(t) = 0. For this to
happen, and u not be identically zero, we require that F (0) = 0. Similarly,
we need F (L) = 0. Consequently, all-together, the function F (x) must
satisfy
c2 F ′′ (x) = λF (x) , (7.55)
where
F (0) = 0 and F (L) = 0, (7.56)
The only difference between the above BVP, and the one for the diffu-
sion equation, is that we now have the coefficient c2 instead of D. Conse-
quently, from (7.20) and (7.21), the nonzero solutions of (7.55) and (7.56)
are  nπx 
Fn (x) = b̄n sin , (7.57)
L
7.5. Wave Equation 201

and  nπ 2
λn = −c2 , (7.58)
L
for n = 1, 2, 3, . . .. Also, b̄n is an arbitrary constant.

Finding G(t)

Now that we know λ, (7.54) takes the form

 nπ 2
G ′′ (t) = −c2 G(t)
L
Assuming that G(t) = ert , we get that r2 = −(cnπ/L)2 . So, r =
±icnπ/L, and from this we get the general solution
Gn (t) = an cos(ωn t) + bn sin(ωn t), (7.59)
where
cnπ
,
ωn = (7.60)
L
and an and bn are arbitrary constants.

The General Solution

We have shown that for any given n, the function un (x, t) = Fn (x)Gn (t)
is a solution of the PDE that satisfies the boundary conditions. The re-
sulting general solution, that satisfies the PDE and boundary conditions,
is, therefore,
X∞
u(x, t) = un (x, t),
n=1
or equivalently
∞
X    nπx 
u(x, t) = an cos(ωn t) + bn sin(ωn t) sin , (7.61)
L
n=1

where an and bn are arbitrary constants, and ωn is given in (7.60). In

writing this down, the constant b̄n in (7.57) has been absorbed into the
an and bn .

Satisfying the Initial Conditions

u(x, 0) = g(x): We need

∞
X  nπx 
an sin = g(x). (7.62)
L
n=1

From (7.31), this means that

2 L  nπx 
Z
an = g(x) sin dx. (7.63)
L 0 L
202 Chapter 7. Partial Differential Equations

ut (x, 0) = h(x): From (7.61), it is required that

∞
X  nπx 
ωn bn sin = h(x). (7.64)
L
n=1

Letting Bn = ωn bn , then the above equation takes the form

∞
X  nπx 
Bn sin = h(x). (7.65)
L
n=1

This is the same problem we had in Section 7.3.5, except that the
coefficient is being denoted as Bn instead of bn . So, from (7.25),
L
2  nπx 
Z
Bn = h(x) sin dx.
L 0 L

Since bn = Bn /ωn , the conclusion is that

L
2  nπx 
Z
bn = h(x) sin dx. (7.66)
cnπ 0 L

7.5.1 Examples
Example 1: Suppose that c = 1, L = 2, g(x) = 3 sin(πx), and h(x) = 0.
In this case, from (7.60), ωn = nπ/2. The resulting general solution
(7.61) is
∞ h
X nπ
nπ
i  nπx 
u(x, t) = an cos t + bn sin t sin .
2 2 2
n=1

To satisfy the initial condition, since h(x) = 0 then, from (7.66),

the bn ’s are all zero. As for the an ’s, note that g(x) is one of the
sine functions in the series. Namely, it is the one when n = 2. This
enables us to avoid the integral in (7.63). To satisfy (7.62) we simply
take a2 = 3, and all the other bn ’s are zero. Therefore, the solution
is
u(x, t) = 3 cos(πt) sin(πx). (7.67)
This solution is shown in Figure 7.9, both as time slices and as
the solution surface for 0 ≤ t ≤ 3T , where T = 2 is the period of
oscillation. 

Example 2: Suppose that in the previous example, the initial conditions

are g(x) = 0 and
 πx   3πx 
h(x) = 3 sin − 4 sin + 5 sin(2πx).
2 2
7.5. Wave Equation 203

3
Solution

-3
0 0.5 1 1.5 2
x-axis

Figure 7.9. Solution of the wave equation in Example 1. Shown is the solution
surface as well as the solution profiles at specific time values.

This consists of the sum of three of the sine functions in (7.65):

n = 1, n = 3, and n = 4. To satisfy (7.65) we take B1 = 3,
B3 = −4, B4 = 5, and all the other Bn ’s are zero. With this,
b1 = 3/ω1 = 6/π, b3 = −4/ω3 = −8/(3π), b4 = 5/ω4 = 5/(2π).
Also, since g(x) = 0, then from (7.63), all the an ’s are zero. The
resulting solution is
6  πt   πx  8  3πt   3πx 
u(x, t) = sin sin − sin sin
π 2 2 3π 2 2
5
+ sin(2πt) sin(2πx). 
2π
7.5.2 Natural Modes and Standing Waves
The curves shown in the lower plot in Figure 7.9 resemble what you see for
time lapse photographs of a vibrating string. There is a reason for this,
which is that the wave equation can be used to model the vibrational
motion of an elastic string. To pursue this a bit further, we found that
the solution of the wave equation problem consists of the superposition
of functions of the form
   nπx 
un (x, t) = an cos(ωn t) + bn sin(ωn t) sin , (7.68)
L
204 Chapter 7. Partial Differential Equations

where
cnπ
ωn = . (7.69)
L
The expression in the square brackets is a periodic function of t, with
period 2π/ωn . In this context, sin(nπx/L) is called a natural mode for
the problem, having natural frequency ωn . The resulting solution in (7.68)
corresponds to what is called a standing wave. So, the curves shown in
the lower plot in Figure 7.9 are plots of a standing wave in the case of
when n = 2. It is also possible to have traveling wave solutions, similar
to waves on a lake or ocean. If you want to learn about traveling waves,
you might look at Strauss [2007] or Holmes [2019].

Exercises
1. You are to find the solution of the wave equation problem for the
following initial conditions. Assume that L = 1 and c = 4. Note that
you should be able to answer this question without using integration.
a) g(x) = sin(3πx), and h(x) = 0
b) g(x) = 0, and h(x) = −2 sin(8πx)
c) g(x) = − sin(πx) + 4 sin(3πx), and h(x) = −3 sin(5πx)
d) g(x) = 5 sin(7πx), and h(x) = 2 sin(8πx) + 3 sin(12πx)
e) g(x) = 2 sin(2πx) cos(πx), and h(x) = −2 sin(8πx)
f) g(x) = 3 cos(2πx − π2 ), and h(x) = −3 cos(7πx) sin(2πx)
2. Find the general solution of the following.
a) uxx = utt , for 0 < x < 1, with the boundary conditions u(0, t) = 0
and ux (1, t) = 0.
b) 4uxx = utt , for 0 < x < 1, with the boundary conditions ux (0, t) = 0
and u(1, t) = 0.
c) uxx = 4utt , for 0 < x < 1, with the boundary conditions u(0, t) =
u(1, t) and ux (0, t) = ux (1, t).
d) uxx = utt +ut , for 0 < x < 1, with the boundary conditions u(0, t) =
0 and u(1, t) = 0. This is an example of what is called a damped
wave equation.
3. Solve
∂2u ∂2u

0 < x < 1,
4 2 = 2 , for
∂x ∂t 0 < t,
where u(0, t) = 0, u(1, t) = 0, u(x, 0) = 0, and ut (x, 0) = x(1 − x).

7.6 Inhomogeneous Boundary Conditions

Solving the diffusion and wave equations using separation of variables
required the boundary conditions to be homogeneous. We now consider
7.6. Inhomogeneous Boundary Conditions 205

how to find the solution when the boundary conditions are inhomoge-
neous, and have the form
u(0, t) = α, (7.70)
and
u(L, t) = β, (7.71)
where α and β are constants. The method used to find the solution is to
write it as
u(x, t) = w(x) + v(x, t),
where we pick w(x) so it satisfies the given boundary conditions. In other
words, so that w(0) = α and w(L) = β. Pretty much any smooth function
can be used, but it makes things easier if w comes from the steady state
equation. What this entails is explained below.

7.6.1 Steady State Solution

The steady state problem is the one that comes from the PDE and bound-
ary conditions when assuming the solution is independent of t. Assuming
we are solving the diffusion equation (7.9), then we are looking for the
function w(x) that satisfies

d2 w
= 0, for 0 < x < L,
dx2
where, from (7.70) and (7.71), w(0) = α and w(L) = β. The resulting
solution is
β−α
w(x) = α + x.
L

7.6.2 Transformed Problem

Now that we know the steady state solution, we write the solution of the
original diffusion problem as
β−α
u(x, t) = α + x + v(x, t). (7.72)
L
Since uxx = vxx and ut = vt , then from the diffusion equation (7.9) we
have that
∂2v

∂v 0 < x < L,
D 2 = , for (7.73)
∂x ∂t 0 < t.
At x = 0, from (7.72), v(0, t) = u(0, t) − α = 0. This also happens at the
other endpoint. So, the boundary conditions are

v(0, t) = 0, (7.74)

and
v(L, t) = 0. (7.75)
206 Chapter 7. Partial Differential Equations

Finally, if the initial condition is u(x, 0) = g(x), then the resulting initial
condition for v is
β−α
v(x, 0) = g(x) − α − x, for 0 < x < L. (7.76)
L
The above problem for v(x, t) has the same form as the one for u(x, t),
as given in (7.9)-(7.12), except for a slightly different looking initial condi-
tion. Consequently, we can use the solution as given in (7.23) and (7.25)
if we make the appropriate adjustments. In particular,
X∞  nπx 
v(x, t) = bn eλn t sin ,
L
n=1
where
L
2 β − α   nπx 
Z
bn = g(x) − α − x sin dx, (7.77)
L 0 L L
and  nπ 2
λn = −D .
L

7.6.3 Summary
We have shown that the solution of
∂2u

∂u 0 < x < L,
D 2 = , for
∂x ∂t 0 < t,
where u(0, t) = α, u(L, t) = β, and u(x, 0) = g(x), is
∞
β−α X  nπx 
u(x, t) = α + x+ bn eλn t sin , (7.78)
L L
n=1

where bn is given in (7.77).

Example: Find the solution of


0 < x < 5,
4∂x2 u = ∂t u , for
0 < t,
where u(0, t) = 3, u(5, t) = 2, and u(x, 0) = 0.
Answer: In this problem D = 4, L = 5, α = 3, and β = 2. So, from
(7.77),
2 5 x   nπx  2 
Z
3 − 2(−1)n .

bn = −3+ sin dx = −
5 0 5 5 nπ
Therefore, from (7.78), the solution of the diffusion problem is
∞
x X 2   nπx 
3 − 2(−1)n eλn t sin

u(x, t) = 3 − − ,
5 nπ 5
n=1

2
where λn = −4 nπ 5 . 
Exercises 207

7.6.4 Wave Equation

The method works, without change, on the wave equation. The only
complication is, as it usually is with the wave equation, differentiability.
To explain, if the boundary conditions are u(0, t) = α and u(L, t) = β,
and the initial conditions are u(x, 0) = g(x) and ut (x, 0) = h(x), then it
is required that
g(0) = α, h(0) = 0, g ′′ (0) = 0,
and
g(L) = β, h(L) = 0, g ′′ (L) = 0.
These are called compatibility conditions. If they are satisfied, and g ′′ (x)
and h′ (x) are continuous, then the solution has the differentiability re-
quited to satisfy the wave equation.

Exercises
1. You are to find the solution of the diffusion equation 4uxx = ut for the
given boundary and initial conditions. Assume that L = 1.
a) u(0, t) = 1, u(1, t) = −1, and u(x, 0) = 0.
b) u(0, t) = 2, u(1, t) = −5, and u(x, 0) = 2.
c) u(0, t) = −4, u(1, t) = 1, and u(x, 0) = x.
2. Find the steady state solution of the following problems.
a) uxx = ut , for 0 < x < 2, with the boundary conditions u(0, t) = 1
and ux (2, t) = −1.
b) 4uxx = ut , for 0 < x < 4, with the boundary conditions ux (0, t) = 2
and u(4, t) = 1.
c) (1 + t)uxx = ut , for 0 < x < 1, with the boundary conditions
u(0, t) = −1 and u(1, t) = 2.
d) uxx = ut +u, for 0 < x < 3, with the boundary conditions u(0, t) = 1
and u(3, t) = 2.
e) uxx −ux = ut , for 0 < x < 2, with the boundary conditions u(0, t) =
−1 and u(2, t) = 1.
3. Solve
∂2u

∂u 0 < x < 2,
= , for
∂x2 ∂t 0 < t,
where u(0, t) = 1, ux (2, t) = −1, and u(x, 0) = 0.
4. Solve 
2 0 < x < 1,
(1 + t)∂x u = ∂t u , for
0 < t,
where u(0, t) = −1, u(1, t) = 2, and u(x, 0) = 0.
208 Chapter 7. Partial Differential Equations

5. Solve
∂2u ∂2u

0 < x < 1,
9 2 = 2 , for
∂x ∂t 0 < t,
where u(0, t) = 1, u(1, t) = −1, u(x, 0) = 1 − 2x − 7 sin(3πx), and
ut (x, 0) = 0.

7.7 Inhomogeneous PDEs

It is common in applications to have a PDE that is not homogeneous. To
explain how to solve such a problem, suppose the PDE is

∂2u

∂u 0 < x < L,
D 2 = + p(x, t) , for (7.79)
∂x ∂t 0 < t,

where p(x, t) is a given smooth function of x and t. It is assumed the

boundary conditions are homogeneous, and so,

u(0, t) = 0, (7.80)

and
u(L, t) = 0. (7.81)

Sine Series Expansions

The general solution when p ≡ 0, which is given in (7.23), consists of

the superposition of functions containing sin(nπx/L). The solution for
nonzero p can also be expanded in this way. Specifically, we can write
∞
X  nπx 
u(x, t) = wn (t) sin , for 0 ≤ x ≤ L, (7.82)
L
n=1

where the wn (t)’s are determined from the PDE. The expansion in (7.82)
is guaranteed from the Sine Convergence Theorem (page 189) because u is
a smooth function and it satisfies the homogeneous boundary conditions
(7.80) and (7.81).
We will also expand the forcing function p is a sine series, and write
∞
X  nπx 
p(x, t) = pn (t) sin , for 0 < x < L, (7.83)
L
n=1

where
L
2  nπx 
Z
pn (t) = p(x, t) sin dx. (7.84)
L 0 L
Because p(x, t) is known, the pn (t)’s are known. Note that it is not
assumed that p = 0 at the endpoints, which is why the interval in (7.83)
is 0 < x < L and not 0 ≤ x ≤ L.
7.7. Inhomogeneous PDEs 209

Solving the PDE

Assuming the series for u can be differentiated term-by-term, we get that

∞  ∞
X nπ 2  nπx  X  nπx 
uxx = − wn (t) sin and ut = wn′ (t) sin .
L L L
n=1 n=1

Introducing these into (7.79), as well as using (7.83), we have that

∞    
X nπ 2 ′ nπx 
D wn (t) + wn (t) + pn (t) sin = 0. (7.85)
L L
n=1

For this to hold, the term in the square bracket must be zero. The proof
of this uses the integration formula (7.30), in exactly the same way it was
used to find the coefficients of the sine series. The conclusion is that
 nπ 2
D wn (t) + wn′ (t) + pn (t) = 0,
L
or equivalently,
wn′ + κn wn = −pn , (7.86)
where  nπ 2
κn = D .
L
This is a first-order linear differential equation for wn , which can be solved
using an integrating factor. The integrating factor in this case is, from
(2.18), µ = eκn t . So, from (2.21), we get the general solution of (7.86) is
 Z t 
−κn t κn s
wn (t) = e − pn (s)e ds + wn (0) . (7.87)
0

Satisfying the Initial Condition

To solve the problem it remains to satisfy the initial condition

u(x, 0) = g(x), for 0 < x < L. (7.88)

From (7.25), and since bn = wn (0), it is required that

2 L  nπx 
Z
wn (0) = g(x) sin dx.
L 0 L

7.7.1 Summary
To summarize our findings, the solution of the inhomogeneous diffusion
problem (7.79)-(7.81), which satisfies the initial condition (7.88), is
∞
X  nπx 
u(x, t) = wn (t) sin , (7.89)
L
n=1
210 Chapter 7. Partial Differential Equations

where  Z t 
−κn t κn s
wn (t) = e − pn (s)e ds + wn (0) , (7.90)
0
L
2  nπx 
Z
pn (t) = p(x, t) sin dx, (7.91)
L 0 L
L
2  nπx 
Z
wn (0) = g(x) sin dx,
L 0 L
and κn = D(nπ/L)2 .

Example
Suppose the problem to solve is

∂2u

∂u 0 < x < 1,
4 2 = + 3 sin(2t) sin(πx) , for (7.92)
∂x ∂t 0 < t,

where
u(0, t) = 0, (7.93)
u(1, t) = 0, (7.94)
and
u(x, 0) = 0, for 0 < x < 1. (7.95)
In this problem, D = 4 and L = 1. The first step is to find the pn ’s. From
(7.91), we want
∞
X
3 sin(2t) sin(πx) = pn (t) sin(nπx).
n=1

So, p1 (t) = 3 sin 2t and all the other pn ’s are zero. Also, since g(x) = 0
then wn (0) = 0, for all n This leaves the integral in (7.90), and so
Z t Z t
κ1 s
p1 (s)e ds = 3 sin(2s)eκ1 s ds
0 0
2 + κ1 eκ1 t sin(2t) − 2eκ1 t cos(2t)
=3 .
κ21 + 4

Since κ1 = 4π 2 , then
3 h
2 −4π 2 t
i
w1 (t) = cos(2t) − 2π sin(2t) − e .
2(1 + 4π 4 )

Therefore, the solution of the diffusion problem is

3 h
2 −4π 2 t
i
u(x, t) = cos(2t) − 2π sin(2t) − e sin(πx). 
2(1 + 4π 4 )
Exercises 211

7.7.2 A Very Useful Observation

As you might have noticed, the problem was solved without using separa-
tion of variables. Instead we assumed that the solution can be expanded
in a sine series, as expressed in (7.82). For this to work it is essential that
the functions sin(nπx/L) satisfy the boundary conditions, which they do
for this problem. By using this sine series expansion, the problem reduces
to solving a relatively simple ODE for the coefficients of the series. This
approach can be used on other PDEs and examples of how this is done
are given in Exercises 3 and 4. In fact, this idea is the basis for what is
called the Galerkin method for computing the solution of a PDE.

Exercises
1. You are to find the solution of the diffusion equation (7.79), where
u(0, t) = 0, u(1, t) = 0, u(x, 0) = 0, and p(x, t) is given below. Assume
that D = 4 and L = 1.
a) p(x, t) = −4 cos(t) sin(5πx).
b) p(x, t) = e−2t sin(3πx).
c) p(x, t) = 1.
2. There is a simpler way to solve an inhomogeneous PDE when the
forcing function does not dependent on t. In this problem assume that
p(x, t) = x2 .
a) Find the steady state solution of (7.79), that satisfies (7.80) and
(7.81).
b) Letting u(x, t) = w(x) + v(x, t), where w(x) is the steady state
solution you found in part (a), find the PDE and boundary condi-
tions satisfied by v(x, t). Also, if u(x, 0) = g(x), then what is the
resulting initial condition for v(x, t)?
c) Assuming g(x) = 0, find v(x, t), and from this determine the solu-
tion of the original diffusion problem.
3. This exercise considers how to use a sine series to solve
∂2u

∂u 0 < x < 2,
2
= + 5u , for
∂x ∂t 0 < t,
where u(0, t) = 0, u(2, t) = 0, and u(x, 0) = x. This is going to be
done using the assumption in (7.82), which for this problem is
∞
X  nπx 
u(x, t) = wn (t) sin , for 0 ≤ x ≤ 2.
2
n=1

a) Assuming u(x, t) is smooth, explain why the Sine Series Conver-

gence Theorem guarantees that the above series converges to u(x, t).
212 Chapter 7. Partial Differential Equations

b) Substitute the series into the PDE and rewrite the result so it re-
sembles (7.85). From this determine the differential equation wn (t)
satisfies.
c) Find the general solution for wn (t), and from this write down the
general solution for u(x, t).
d) Use the general solution to satisfy the initial condition, and from
this determine the solution of the problem.
4. Solve
∂2u

∂u 0 < x < 3,
(1 + t) = , for
∂x2 ∂t 0 < t,
where u(0, t) = 0, u(3, t) = 0, and u(x, 0) = 1. Find the solution using
the procedure outlined in Exercise 3 (with the appropriate modifica-
tions).

7.8 Laplace’s Equation

We are going to consider how to solve the equation

∇2 u = 0, (7.96)

where ∇2 is called the Laplacian, or the Laplacian operator. The

formula for ∇2 depends on the coordinate system you are using. In the
case of Cartesian coordinates,
∂2 ∂2
∇2 = + ,
∂x2 ∂y 2
in which case (7.96) is simply

uxx + uyy = 0. (7.97)

Later we will consider polar coordinates, and the respective formula for
∇2 will be given at that time.
It should not be a surprise that (7.96) is known as Laplace’s equa-
tion. It plays a fundamental role in applied mathematics. If you look
through a junior or senior level textbook in complex variables, fluid dy-
namics, electromagnetism, heat transfer, etc, it will appear often. As
an example, heat conduction is governed by the diffusion equation ut =
D∇2 u. So, if you want to determine the steady-state temperature distri-
bution, then you must solve (7.96).
Our goal is to find the function u(x, y) that satisfies Laplace’s equation
for (x, y) in a region, as illustrated in Figure 7.10, along with a boundary
condition u = f on the boundary of the region. To keep things simple we
will only consider simple shapes, and that means rectangular and circular.
In both cases, the method of separation of variables is used to find the
solution.
7.8. Laplace’s Equation 213

Figure 7.10. The solution u(x, y) is to satisfy Laplace’s equation in a given

region in the x,y-plane, and also satisfy u = f on the boundary of the region.

The symbol ∇2 appearing in Laplace’s equation comes from vector

calculus. Namely, using the gradient ∇ and the dot product, one writes

∇2 = ∇ · ∇.
∂ ∂


In Cartesian coordinates the gradient is ∇ = ∂x , ∂y , and from this you
∂2 ∂2
get that ∇2
= +
∂x2 ∂y 2
.
This can also be used to derive the formula for
2
∇ is other coordinate systems, such as polar coordinates.

7.8.1 Rectangular Domain

The problem to solve is

0 < x < a,
uxx + uyy = 0 , for (7.98)
0 < y < b,

where the boundary conditions are shown in Figure 7.11. So, u = 0 when
x = 0, when x = a, and when y = 0. Along the top, where y = b,
u = f (x).
The steps used in carrying out the separation of variables method
are very similar to what was done earlier. We will first find the general
solution of the PDE that satisfies the homogeneous boundary conditions.

Figure 7.11. Rectangular domain used when solving Laplace’s equation and
the corresponding boundary conditions.
214 Chapter 7. Partial Differential Equations

We will then use that solution to satisfy the inhomogeneous boundary

condition (which is on the upper side of the rectangle).

Separation of Variables Assumption

Assuming
u(x, y) = X(x)Y (y), (7.99)

and then substituting this into Laplace’s equation gives us

X ′′ (x) Y ′′ (y)
=− . (7.100)
X(x) Y (y)

Since the left-hand-side is only a function of x, and the right-hand-side is

only a function of y, we can conclude that there is a constant λ so that

X ′′ (x) = λX(x) , (7.101)

and
Y ′′ (y) = −λY (y) . (7.102)

As explained earlier, the separated solution (7.99) is required to satisfy

the homogeneous boundary conditions shown in Figure 7.11.

Finding X(x) and λ

The equation to solve is

X ′′ (x) = λX(x) . (7.103)

Since u = 0 when x = 0 and x = a, then it is required that

X(0) = 0 and X(a) = 0. (7.104)

This is essentially the same problem we had when solving the diffusion
and wave equations, and the general solution is
 nπx 
Xn (x) = cn sin , (7.105)
a

and
 nπ 2
λn = − , (7.106)
a
for n = 1, 2, 3, . . .. Also, cn is an arbitrary constant.
7.8. Laplace’s Equation 215

Finding Y (y)

From (7.102), we now need to solve

Y ′′ (y) = −λn Y (y) ,

where Y (0) = 0. The general solution of this ODE is

Yn = An enπy/a + Bn e−nπy/a .

To have Yn (0) = 0 we need Bn = −An . Consequently,

Yn = An enπy/a − e−nπy/a

= 2An sinh(nπy/a). (7.107)

The General Solution

The resulting general solution, that satisfies the PDE and homogeneous
boundary conditions, is, therefore,
∞
X
u(x, y) = Xn (x)Yn (y),
n=1

or equivalently
∞
X  nπy   nπx 
u(x, y) = cn sinh sin , (7.108)
a a
n=1

where the cn ’s are arbitrary constants. In writing this down, the constant
2An in (7.107) has been absorbed into the cn in (7.105).

Satisfying the Inhomogeneous Boundary Condition

To have u(x, b) = f (x), we need

∞
X  nπb   nπx 
cn sinh sin = f (x). (7.109)
a a
n=1

This can be written as

∞
X  nπx 
bn sin = f (x),
a
n=1

where bn = cn sinh(nπb/a). According to the Sine Convergence Theorem

(page 189), the bn ’s that satisfy the above equation are

2 a  nπx 
Z
bn = f (x) sin dx.
a 0 a
216 Chapter 7. Partial Differential Equations

From this we conclude that

a
2  nπx 
Z
cn = f (x) sin dx. (7.110)
a sinh(nπb/a) 0 a

With this value for cn , u(x, y) given in (7.108) is the solution of the
problem.

Example 1: Find the solution of


0 < x < 1,
uxx + uyy = 0 , for
0 < y < 1,

where u(x, 1) = 8 sin(5πx) and u = 0 on the other three sides of the

square (see Figure 7.11).
Answer: In this problem a = b = 1, and f (x) = 8 sin(5πx). From
(7.109), we need
∞
X
8 sin(5πx) = cn sinh(nπ) sin(nπx).
n=1

So, c5 sinh(5π) = 8, and the other cn ’s are zero. Therefore, the

solution is
sinh(5πy)
u(x, y) = 8 sin(5πx).
sinh(5π)
The resulting solution is shown in Figure 7.12. 

Figure 7.12. Solution of Laplace’s equation derived in Example 1.

7.8. Laplace’s Equation 217

7.8.2 Circular Domain

We now solve Laplace’s equation when the domain is the circular region
x2 + y 2 < a2 , as illustrated in Figure 7.13. It makes it easier in this
case to use polar coordinates and take x = r cos θ, y = r sin θ. Using the
standard change of coordinate formulas, one finds that

∂2 1 ∂ 1 ∂2
∇2 = + + .
∂r2 r ∂r r2 ∂θ2
Therefore, the problem we are solving is

1 1 0 ≤ r < a,
urr + ur + 2 uθθ = 0 , for (7.111)
r r 0 ≤ θ < 2π,

where the boundary condition is

u r=a
= f (θ). (7.112)

As will be seen below, this problem is easily solved using separation of

variables.
Using polar coordinates makes solving the problem easier, but it re-
quires some comment. First, (7.111) is singular when r = 0. This always
happens when using polar coordinates. To prevent the singular nature of
the equation from interfering with us solving the problem, it is assumed
that the solution is bounded. The second comment is that the positive
x-axis corresponds to θ = 0 and to θ = 2π. The solution u and its deriva-
tive uθ must be continuous in the circular domain, and this means we
must require that

u θ=0
=u θ=2π
and uθ θ=0
= uθ θ=2π
. (7.113)

In the vernacular of the subject, these are called periodic boundary con-
ditions. Also, note that these boundary conditions are homogeneous be-
cause u = 0 satisfies both of them. Finally, if (7.113) hold then ur is also
continuous in the domain.

Figure 7.13. Circular domain and corresponding boundary condition.

218 Chapter 7. Partial Differential Equations

Separation of Variables Assumption

Assuming
u = R(r)Θ(θ), (7.114)
and then substituting this into Laplace’s equation (7.111) gives us

R ′′ (r) R ′ (r) Θ ′′ (θ)

r2 +r =− . (7.115)
R(r) R(r) Θ(θ)
Since the left-hand-side is only a function of r, and the right-hand-side is
only a function of θ, we can conclude that there is a constant λ so that

r2 R ′′ (r) + rR ′ (r) = λR(r) , (7.116)

and
Θ ′′ (θ) = −λΘ(θ) . (7.117)
From (7.113) we also must have that

Θ(0) = Θ(2π) and Θ ′ (0) = Θ ′ (2π). (7.118)

Finding Θ(θ) and λ

As usual, the first problem to solve is the one involving the homogeneous
boundary conditions, which means solving (7.117) and (7.118).

λ = 0 : In this case (7.117) is Θ ′′ = 0, and so Θ = A + Bθ, where A and

B are constants. To satisfy (7.118) it must be that B = 0, and so
the solution is Θ0 = A0 .

λ 6= 0 : Assuming Θ = erθ , we get the characteristic equation r2 = −λ.

Skipping the two real roots case we take λ > 0 , giving us the general
solution √ √
Θ = A cos(θ λ) + B sin(θ λ).
√
To√satisfy (7.118) one finds that cos(2π λ) = 1. This means that
2π λ = 2π, 4π, 6π, . . .. In other words,

λn = n2 , for n = 1, 2, 3, . . . , (7.119)

and
Θn = An cos(nθ) + Bn sin(nθ). (7.120)

Finding R(r)

λ = 0 : In this case (7.116) is r2 R ′′ + rR ′ = 0. This is an Euler equation,

and how to solve it was explained in Section 3.11. One finds that
R = A + B ln r, where A and B are constants. To have a bounded
solution we require B = 0, and this means the solution is R0 = A0 .
7.8. Laplace’s Equation 219

λ = n2 : Now (7.116) is
r2 R ′′ + rR ′ = n2 R.
This is also an Euler equation, and the general solution is Rn =
An rn + B n r−n . Because the solution is bounded we require B n = 0.

The General Solution

The general solution of Laplace’s equation that is bounded and satisfies

(7.113) is
∞
X
u= Rn (r)Θn (θ),
n=0

or equivalently
∞
1 X 
rn an cos(nθ) + bn sin(nθ) .

u = a0 + (7.121)
2
n=1

The coefficients in this formula are written in a form similar to what was
used earlier for a sine and cosine series. So, for example, we have written
R0 Θ0 = A0 A0 = 21 a0 .

Satisfying the Boundary Condition

To have u = f (θ) when r = a, we need

∞
1 X 
an an cos(nθ) + bn sin(nθ) = f (θ).

a0 + (7.122)
2
n=1

The an ’s and bn ’s are determined in the same way as for a sine and cosine
series. For example, to determine a7 you multiply the above equation by
cos(7θ), integrate for 0 ≤ θ ≤ 2π, and use orthogonality conditions such
as given in (7.30) and (7.39). The resulting formulas obtained in this way
are Z 2π
1
an = f (θ) cos(nθ)dθ,
πan 0
and
2π
1
Z
bn = f (θ) sin(nθ)dθ.
πan 0

Example 2: Find the solution of

∇2 u = 0 , for x2 + y 2 < 1,

where u = 3 sin(4θ) for x2 + y 2 = 1.

220 Chapter 7. Partial Differential Equations

Answer: In this problem a = 1 and f (θ) = 3 sin(4θ). From (7.122),

we need
∞
1 X 
3 sin(4θ) = a0 + (an cos(nθ) + bn sin(nθ) .
2
n=1

So, b4 = 3, and the other coefficients are zero. Therefore, the solu-
tion is
u = 3r4 sin(4θ).
The resulting solution is shown in Figure 7.14. 

Exercises
1. You are to find the solution of the problem shown in Figure 7.11.
Assume that a = 1 and b = 2. Note that you should be able to answer
this question without using integration.
a) f (x) = 5 sin(2πx)
b) f (x) = −3 sin(12πx)
c) f (x) = sin(πx) − 7 sin(8πx)
d) f (x) = −3 sin(4πx) − sin(7πx) + 6 sin(20πx)
2. You are to find the solution of the problem shown in Figure 7.13.
Assume that a = 2. Note that you should be able to answer this
question without using integration.
a) f (θ) = 4 cos(3θ)

Figure 7.14. Solution of Laplace’s equation derived in Example 2.

Exercises 221

b) f (θ) = 1 − 3 sin(15θ)
c) f (θ) = sin(θ) + 3 cos(5θ)
d) f (θ) = 4 − 2 sin(5θ) − 4 sin(9θ) + 8 cos(14θ)

Figure 7.15. Problem solved in Exercise 3.

3. The problem concerns solving the problem shown in Figure 7.15.

a) Write down the differential equation and boundary conditions for
this problem.
b) Find the general solution. This should satisfy Laplace’s equation as
well as the homogeneous boundary conditions.
c) Use the inhomogeneous boundary condition to find the formula for
the coefficient in your general solution.
d) If g(y) = 7 sin(3πy), then what is the solution?
e) If g(y) = −2 sin(2πy) + 8 sin(7πy), then what is the solution?
4. The problem concerns solving the problem in the quarter circle shown
in Figure 7.16.
a) In polar coordinates, write down the differential equation and bound-
ary conditions for this problem.
b) Find the general solution. This should satisfy Laplace’s equation
as well as the homogeneous boundary conditions. It should not be

Figure 7.16. Problem solved in Exercise 4.

222 Chapter 7. Partial Differential Equations

required to satisfy the conditions in (7.113), as those only apply

when you have a domain with 0 ≤ θ ≤ 2π.
c) If f (θ) = −3 sin(4θ), then what is the solution?
d) If f (θ) = 9 sin(2θ) − 5 sin(14θ), then what is the solution?
5. Setting λ = −κ2 , where κ > 0, what is the general solution of (7.117)?
Show that to satisfy the boundary conditions (7.118) that the solution
is identically zero.
6. Suppose that instead of using 0 ≤ θ < 2π, one were to use −π ≤ θ < π.
How are the periodic boundary conditions (7.113) changed?
 Appendix A

Matrix Algebra: Summary

The following is a brief summary of the rules of matrix and vector algebra
in two dimensions.

A.1 Addition: x + y and A + B

General:
           
x u x+u a b e f a+e b+f
+ = + =
y v y+v c d g h c+g d+h

Examples:
           
1 −3 −2 1 −2 1 0 2 −2
+ = + =
2 4 6 −3 4 −7 3 1 7

A.2 Scalar Multiplication: αx and αA

Scalar means a number (real or complex).
General:
           
x x αa a b a b αa αb
α = α= α = α=
y y αb c d c d αc αd

Examples:
       
1 4 1 0 −4 0
4 = −4 =
−3 −12 −2 3 8 −12

       
4 1 6 0 2 0
=4 =3
−8 −2 3 −12 1 −4

223
224 Appendix A. Matrix Algebra: Summary

A.3 Equality: x = y and A = B

General:    
x u
= means that x = u, y = v
y v
   
a b e f
= means that a = e, b = f, c = g, d = h
c d g h

Examples: note that I is defined on page 85

 
a
= 0 means that a = 0, b = 0
b

 
a b
=I means that a = 1, b = 0, c = 0, d = 1
c d

A.4 Matrix-Vector Multiplication: Ax and αAx + y

General:     
a b x ax + by
=
c d y cx + dy

Examples:     
1 −2 1 −3
=
−3 4 2 3
          
1 −1 −1 1 −2 1 −7
3 − =3 − =
0 2 1 5 2 5 1

A.5 Differentiation: x′ and (af (t))′

General:
x′ (t)
   
d x(t)
=
dt y(t) y ′ (t)
    
d a a ′
f (t) = f (t)
dt b b

Examples:
t + t3 1 + 3t2
   
d
=
dt sin t cos t
      
d 1 5t 1 5t 1
e = 5e = 5 e5t
dt −2 −2 −2
Appendix B

Answers

Chapter 1
Section 1.2, pg 4
2a) r = −2 2g) none 3c) r = 1/3, c = 3
2b) r = 1/3 2h) r = 0 3d) r = 1, c = −1
2c) none 2i) none 3e) r = −2/5, c = −7
2d) r = 0, −4 2j) none 3f) r = −4, c = 3
2e) r = −3, 1/2 3a) r = −2, c = 1
2f) r = 2 3b) r = −1, c = −1

Chapter 2
Section 2.1, pg 12
 2

1a) y = (9t + c)−1/3 and y = 0 2e) y (t) = ln 1/2 t e+2
e
1b) y = ±(2e−t + c)−1/2 and y = 0 √
2f) y (t) = −2 + 4 + 2 t
1c) y = −1/(cos
p t + c) and y = 0 2g) y (t) = 2 arctan (1 + t)
1d) y = 3 ± t2 /2 + c
−1
2h) y (t) = 5 1 + 4 e5 t
1e) y = − ln( 12 t2 + 2t + c)
2i) y (t) = 21 ln e−2 t + e2 − 1 + t


1f) y = − 31 ln[3 ln(t + 1) + c] √

1g) y = − 41 ln(2e2t + c) 2j) y (t) = ln t/2 + 1/2 t2 + 4
1h) y = − ln12 ln[t ln(2) + c] 3a) q (r) = − √141r+1
1i) y = 31 [−1±(6t+c)−1/2 ], y = − 13 3c) h (τ ) = −2 + 4 eτ /3

−1
1j) y = −2 − 1/(t + c) and y = −2 3d) h (x) = 6 2 + e3 x
1k) y = 3 − 2/(t + c) and y = 3 3e) z (r) = 6 (1 + 6 ln ((1 + er
)/2))
−1
1l) y = tan(t/3 + c) 3f) w (τ ) = 1/2 ln 1/8 τ 4 + 1
1m) y = ln[tan(t2 /2 + c)] 2
t 3g) r (θ) = 2 (θ +√1)
1n) y = ln(ce
√ − 1)
3h) r (θ) = −1 + 2 θ2 + 1
1o) y = ± cet2 − 1
4a) y − ln(1 + y) = t + 1 − ln 2
2a) y (t) = 5 √1501 t+1
4b) 15 t = y 5 + 5 y + 6
2c) y (t) = 4 + 7 t 4c) y + ln(1 + y) = t + 5 + ln(6)
2d) y (t) = (1 + ln (4 + et ) − ln (5))
−1
4c) y − e−y = t + 2 − e−2

Section 2.2, pg 18

225
226 Appendix B. Answers

1a) y (t) = c e−3 t 2e) y (t) = −t+10

5+t
1b) y (t) = −t/2 − 1/4 + e2 t c 2f) y = −(2/3)e
R t s2 /6
−t2 /6
e ds
0
1c) y (t) = −2 t − 14 + et/4 c 3a) q (z) = 2 − 3 e−2 z
1d) y (t) = et − 1 + e−t c 3b) p (x) = −x/4+1/16−1/16 e−4 x
1e) y (t) = 20 t−cos(4
12 t+8
t)+c
3c) w (τ ) = 1/3 e2 τ − 1/3 eτ /2
t+c e4 τ
1f) y (t) = t+2 5
3d) z (τ ) = −τ /4 − 16 + 5 16
Rt√
1g) y = − 13 +e3t 0 se−3s ds+ce 3t 3e) h (x) = −x+14
x+7
3f) h (z) = 35 z−1

t s/2
1h) y = e−t/2 12 0 se
R
1+s ds + c
z+1
4a) yp = −3, yh = ce2 t
2a) y (t) = −4 + 3 et
4b) yp = 3 te−t , yh = ce−t
2b) y (t) = 3/4 t − 3/16 + 3/16 e−4 t
4c) yp = −3 + 1/13 e2 t , yh = cet/7
2c) y (t) = 2 e−t/5 2

2d) y (t) = −1/3 e−t + 2 − 2/3 et/2 4d) yh = ce−t

6b) w (t) = √5−41 e−2 t

Section 2.3, pg 28
ln(2)
1c) ln(4/3) days 8d) 792 + 968e−20/11 ft
2d) either 40 or 39 BC 8e) −22 fps
3c) 50 ln(10) min 9e) m/c − cL/A √
11a) P = N4 4 + z − 8z + z 2 , z = e−rt


4c) 5(1 − e−6 ) g
5b) 104 (1 − e−1 ) kg 9−e−2t
12a) P = 250 3−e−2t
25 11
6c) 324000


11 1 − 27 lbs 13d) 5 ln(64/39)
ln 2 min
7a) v = −20 + 120e−t/2 m/s 14c) k = (4 − 21/4 )/5
1/4
7c) 40(5 − ln 6) m 15b) 120 ln(42/37)
ln(4/3) min
8a) v = −(176/c)(1 − e−2ct/11 ) fps
15c) 120 ln(93/74)
ln(4/3) min

Section 2.4, pg 38
us=unstable; as=asymptotically stable
1a) as 2c) y = ±2, as; y = 0, us
1b) us 2d) y = − ln 2, as
1c) as 2e) y = −2, as; y = 2, us
1d) us 2f) y = 0, as; y = ln 3, us
2a) y = 1, us; y = −2, as 9c) 750
2b) y = ±1, us; y = 0, as

Chapter 3
Section 3.5, pg 50
1a) −7, 1 3d) y (t) = c1 e−t/2 + c2 et/2
1b) −2,√ 2 3e) y = c1 + c2 t
1c) 21 e2 3, 12 e2 3f) y (t) = c1 e3 t + c2 e3 t t
√
1d) 1 + 2e2 3, 1 + 2e 2 3g) y (t) = c1 e−t/2 + c2 e−t/2 t
1
√ 2 1
√
1e) 2 ( 3 − 1)e , 2 ( 3 + 1)e2 3h) y = c1 sin (t/2) + c2 cos (t/2)
1f) −e12 , 0 3i) y (t) = c1 et sin (t) + c2 et cos (t)
3a) y (t) = c1 e−2 t + c2 et 3j) y = e−t (c1 sin (2 t)+c2 cos (2 t))
3b) y (t) = c1 e−2 t + c2 et/2 4a) y (t) = −1/3 e2 t + 1/3 e−t
3c) y (t) = c1 + c2 e−3 t 4b) y (t) = −4/5 et/2 − 1/5 e−2 t
 227

4c) y (t) = −4/3 + 1/3 e−3 t 4i) y = −e−t sin (2 t) − e−t cos (2 t)
4d) y (t) = 4 − 5 et/5 √ 4j) y = 2 et/2 cos (t/3)
3
4e) y = 3 exp(−(1/3) 3t) 6a) (1 + t)
2


4f) y = 5 exp(−(1/2)t) 6b) cos t + 6 t
4g) y (t) = −e−t − e−t t 6c) t + 2
4h) y (t) = −1/3 sin (3 t) − cos (3 t)

Section 3.8, pg 58
1a) y (t) = e3 t c2 + e−2 t c1 − et
2
1b) y = c1 e−2t + c2 e−t + −π sin(πt)−3π cos(πt)+2 sin(πt)
π 4 +5π 2 +4
1c) y (t) = e−5 t c2 + et c1 − 2/5 t2 − 16 t
25 − 125
84
3 sin(2 t) 15 cos(2 t) 15
1d) y (t) = 5 et/5 c1 − 202 − 101 − 101 + 1/6 e−t + c2
15 t2
2t
1e) y (t) = e c2 + e −t/3
c1 − 1/2 t + 4 − 4 + 535
3 89 t
8
1f) y (t) = e−t/4 c2 + c1 et/2 + 4 cos(2 221
t)
− 33 sin(2 t)
221 − 4
1g) y (t) = sin (2 t) c2 + cos (2 t) c1 + 1/25 (5 t − 2) et
1h) y (t) = c2 e−t + c1 e6 t + +1/9 (3 t − 1) cos (3 t) + 1/15 (−5 t − 2) sin (3 t)
1i) y (t) = sin (2 t) et c2 + cos (2 t) et c1 + t2 + 4/5 t + 18 25
1j) y (t) = e−t sin (3 t) c2 + e−t cos (3 t) c1 + 1/10 + 3/13 et
1k) y (t) = 1/3 e3 t c1 − 1/9 t2 − 1/9 t3 − 1/12 t4 + 52 t
27 + c2
2/3 t −2 t 3 t −2 t
1l) y (t) = e c2 + e
−t
c1 − 1/2 e e + 3/8 e
1m) y (t) = e4 t sin (t) c2 + e4 t cos (t) c1 − 1/2 e4 t t cos (t)
1n) y (t) = e6 t c2 + e−t c1 − 15 cos(t+7)
74 + 21 sin(t+7)
74
1o) y (t) = −e−2 t c1 − 3 sin(2 40
t)
+ 1/4 + 1/40 cos (2 t) + e−t c2
1p) y (t) = −4 e−t/4 c1 − 2 sin(2 65
t)
− cos(2 t)
260 + c2
t −2 t
2a) y (t) = e − 1/4 e − 3/2 t − 3/4
2b) y (t) = −1/8 + 9 cos(2 8
t)
+ 1/4 t2
t
2c) y (t) = −1/2 e − 1/2 sin (t) + 1/2 cos (t) + 1
−3 t
2d) y (t) = − 2 e27 + 1/3 t2 − 2/9 t + 29 27
e−2 t
2e) y (t) = 19 16 + 11/4 e−2 t t − 3/16 e2 t
2f) y (t) = 1/2 e−t/2 + 3/4 et/2 − 1 + 1/4 (−t − 1) e−t/2
2g) y (t) = −1/9 sin (3 t) + 1/3 cos (3 t) t
2h) y (t) = −1/2 e−t sin (2 t) − 5/4 e−t cos (2 t) + 1/4 e−t
t/2 t/2
2i) y (t) = 51 e 13 sin(t/2)
+ 21 e 13 cos(t/2)
− 12 sin(3
13
t)
− 34 cos(3
13
t)
t 6t
4a) y (t) = −2/5 e + e c
4b) y (t) = e−2/3 t c + −3 π cos(π t)+2 sin(π t)
9 π 2 +4
4c) y (t) = 2/3 t − 2/9 + e−3 t c
4d) y (t) = −3 t − 15 − 1/6 e−t + et/5 c
4e) y (t) = − 101
t cos (2 t) − 1/25 cos (2 t) − 1/5 t sin (2 t) − 3 sin(2
100
t)
+ e4 t c
6t
4f) y (t) = −1/7 e t − 1/49 e − 1/3 + e c
−t −t

4g) y (t) = −1/10 e−t cos (t) + 3/10 sin (t) e−t + e−2/3 t c
4h) y (t) = −1/4 cos (2 t + 5) + 1/4 sin (2 t + 5) + e2 t c
Section 3.9, pg 63
1a) y = −2 e−2 t + 2 et/2 − 5 e−2 t t
1b) y = 3 + (−3 cos (t) + 3 sin (t)) et
Rt Rt
1c) y = −e−2 t 0 ln (1 + s) e2 s ds + et 0 ln (1 + s) e−s ds
t
1d) y = t/3 + 2/15 t5/2 + e−3 t 0 −1/3 s3/2 + 1 e3 s ds
R

228 Appendix B. Answers

R t −s/5
1e) y = −2 ln (t + 1) + et/5 0 2 e1+s ds
t Rt
1f) − 41 e−t/2 0 sin s2 + 1 es/2 ds + 14 et/2 0 sin s2 + 1 e−s/2 ds
R

3a) 2 t (−t + et − 1)
3b) 1/2 (t − 1) e2 t + 1/2 + t/2
3c) 4 t5/2 √
4b) 1/2 sin (t) t
Section 3.10, pg 73
√ −2t √
2b) u = 14 cos(8t − π) 1
7b) u = − 90 15e sin(2 15t)
2e) 1 √ √ 7d) e−τ /24, τ = z( 3π2 −Arctan(z)),
2 5π

√
3b) u = 3 3 cos 2 3t − 6 z = 1/ 15
√ √
3f) 3π/18√ 8d) u = 2e−3t cos(t − π/4)
1
2 cos 10t − 7π


4b) u = 20 12b) u (t) = 3/4 sin (16 t) t
√ 4
4e) 5(2 + 2); 3π/40 13b) u (t) = 5 sin(3
12
t)t

Section 3.11, pg 77
1a) y (x) = c1 x2 + c2 x2 ln (x)
1b) y (x) = c1 x3 sin √ (ln (x)) + c2 x3 cos (ln (x))
c
1c) y (x) = √1x + c2 3 x
√ √
√ √

1d) y (x) = c1 x sin 1/2 3 ln (x) + c2 x cos 1/2 3 ln (x)
1e) y (x) = c1 x2 sin (3 ln (x)) + c2 x2 cos (3 ln (x))
1f) y (x) = cx1 + xc2/5
2

1g) y (x) = c2 ln (x) + c1

1h) y (x) = c2 x3 + c1
1i) y (x) = xcn1 + c2 xn+2
2a) y (x) = −x2 e + xex
2b) y (x) = −1/4 x4 − x + 1/4 + x2
2c) y (x) = −2 x + x ln (x) + ln (x) + 2
2d) y (x) = 1/4 x2 − 3/2 ln (x) + 3/4
−1
2e) y (x) = 1/2 (x − 1) + x/2 − 1/2

Chapter 4
Section 4.3, pg 91
2a) 3, −2 2b) 5, −1 3a) 2 ± 2i 3b) −1 ± 2i
Section "
4.5, pg 98
c 1 e−3 t + c 2 e2 t
#
1a)
−1/3 c 1 e−3 t + 1/2 c 2 e2 t
c 1 e−t/2 + c 2 et/2
" #
1b)
−2 c 1 e−t/2 + 2 c 2 et/2
c 1 + c 2 e5 t
" #
1c)
5t
" −2 c 1 + 3 c2 t2 e #
−c 2 e
1d)
c 1 e2 t + c 2 e2 t t
c 1 e−2 t
" #
1e)
−2 t
" c 2e #
c 1 sin (3 t) + c 2 cos (3 t)
1f)
−1/3 c 1 cos (3 t) + 1/3 c 2 sin (3 t)
 229

 
c 1 sin (t/2) + c 2 cos (t/2)
1g)  c 1 (−1/5 sin (t/2) + 1/10 cos (t/2)) 
 

+c 2 (−1/5 cos (t/2) − 1/10 sin (t/2))

c 1 et/2 sin (t) + c 2 et/2 cos (t)
 

1h)  c 1 −2 et/2 sin (t) + 4 et/2 cos (t)


 

+c 2 −2 et/2 cos (t) − 4 et/2 sin (t)

√
√

c 1 e2 t sin 2t + c 2 e2 t cos 2t
 
√ √ √
1i)  c 1 1/3 e2 t sin 2t
+ 1/3 e2 t 2 cos 2t


 
2t
√
2t
√ √


 +c 2 1/3 e cos18 2t − 1/3 e 2 sin 2t
 2 a) c 1 = 2/5, c 2 = 5 , b) {c 1 = 7/4, c 2 = 9/4} ,
c) c 1 = 13 5 , c 2 = 7/5 , d) {c 1 = −1, c 2 = −4} , e) {c 1 = 4, c 2 = −1} ,
f ){c 1 = 3, c 2√= 4} , g) {c 1 = −2, c 2 = 4} , h) {c 1 = 7/4, c 2 = 4} ,
i) c 1 =  −7/2 2, c 2−t= 4
c 1 e + c 2 e2 t

 2t

 −2 c 1 e + c 2 e − c 3 e 
4a)  −t −t 

2t
 c 1 te + c 2 2et + c 3 e t 
−t −t

c 1 e + 3c 2 e + 2c 3 e t
 
4b) 
 c 2 e2 t 

2t t
c 2 e +c 3 e
2 c 1 et − c 2 e−2 t
 
 
4c) 
 3 c 1 et 

t −2 t
7c 1 e + 4c 2 e +c 3 e −t

c 1 e−t
 
 √ √ 
4d) 
 2 c 2 e 5t + 2 c 3 e− 5t 

√ √

− 5t √
5t
√

−c 3 5+1 e +c 2 e 5−1
Section 4.7, pg 110
us=unstable; as=asymptotically stable; ns=neutrally stable
1a) us 1d) us 1g) as 3a) us 3d) us
1b) as 1e) as 1h) ns 3b) as
1c) us 1f) us 1i) ns 3c) as

Chapter 5
Section 5.1, pg 115
2a) (u, v) = (1/2, 0), (1/3, 1/4) 2e) (S, I) = (5, 0), (1, 2)
2b) (u, v) = (0, 0), (−1, 1) 2f) (s, c) = (−1, 1)
2c) (u, v) = (1/4, 4) 2g) (x, y) = (0, 0)
2d) (S, P ) = (1, 1), (0, 0), (2, 0) 2h) (x, y) = (0, 2), (0, −1), (2, 0)
Section 5.2, pg 126
1a) (u, v) = (1/2, 0) as; (1/3, 1/4) as
1b) (u, v) = (0, 0) us
1c) (u, v) = (1, −1) us
1d) (u, v) = (1/4, 4) as
230 Appendix B. Answers

1e) (x, y) = (0, 0) as; (x, y) = (2/3, 4/9) us

1f) (x, y) = (0, 0) id
1g) (x, y) = (0, 1) us
1h) (x, y) = (0, 0) us; (x, y) = (c/d, a/b) id
1i) (S, P ) = (0, 0) us; (S, P ) = (2, 0) us; (S, P ) = (1, 1) as
1j) (S, I) = (1, 0) as; (S, I) = (2, −1/2) us
1k) (r, s) = (1, 1) us
1l) (S, E) = (0, E0 ) as
Section 5.3, pg 137
1a) H = v 2 + 3e2u /2 − 3u
1
1b) H = v 2 /2 + 10 ln(1 + 5u2 )
1d) H = 5v 2 /2 + 7u2 /2 + 35 u10

3/2
1d) H = v 2 /2 + 1/3 u2 + 1 3 u2 − 2


p
4d) 3/2
4e) −1
√ R1
4f) 2 2 −1 [(3 + u2 )(1 − u2 )]−1/2 du
√
5d) 2(1 − e−1 )
5e) − ln(2 − e−1 )
√ R1
5f) 2 − ln(2−e−1 ) [(1 − e−1 )2 − (1 − e−u )2 ]−1/2 du

Chapter 6
Section 6.1, pg 149

e−1−s
+ 3 1−es
−1 −s
1a) − (s − 5) 2d) 1+s
1b) 4+3s2
s
2e) 5 s − 1−2se2
−1 −3 s

−1
1c) 2 s−2 + 4 (1 + s) 2f) 2 e
−12 s
−e−15 s
1 s
1d) −9 (s+2)(s−7) 3a) s2s+9
1e) 8 s −3
−1
2 3b) 28 s2 + 49
1f) 9 s −6s3
s+2
3c) (s+1)s+1
2
−2 +π 2
1g) (1 + s) 2
s +3
−3 cos(4)π−sin(4)s 3d) 2 (s2 +9)(s 2 +1)
1h) −2 9 π 2 +s2 s
5 s2 −8 s+4
4a) 6 (s2 +9)2
1i) (s−2)(s2 +4) 2
−2 s 4b) 6 (ss2 +49)
−49
2
2a) 5 e s
−2 s s(s2 −3)
2b) 1−es2 4c) 2 (s2 +1)3
−2 s s+2
2c) − 1−2 e s +e 4d) 10
−s
2
((s+2)2 +25)

Section 6.2, pg 155

−6 s
1a) e s 2a) 2/3 sin (3 t)
1b) es2
−s
2b) 3 te−4 t
−2 s 2c) 1/5 et − 1/5 e−4 t
1c) e s
−2 s
e−5 s
2d) e−t cos (2 t)
1d) 3 e −4 s 2e) 1/4 e2 t + 7/4 e−2 t
−3 s
1e) 4 1−es 2f) 1/3 e−t (6 cos (3 t) − 5 sin (3 t))
1f) 3−e 2g) 1/7 cos √(3
t) − 1/7 cos (4 t)
−s

6s
1g) 1−e +3 es −e
−s −2 s −3 s
2h) cosh 2t − cosh (t)
−2 s −4 s −8 s 2i) sin (2 t) − sin (3 t)
1h) e +e s −e
 231

 
2j) 5 t − 3 e2 t 1
− es
−s
4a) 1−e−2 s s
−1

2k) H (t − 3) e3−t cos (−9 + 3 t)

− e (1−e
s+e−s −1
−s
4b) −s )s2
2l) −1/2 H (t − 2) (t − 2) (t − 4)
1+e−2 s −2 e−s
2m) H (t − 1) − H (t − 2) + H (t − 3) 4c) (1−e−2 s )s2
2n) 2 − t + 2 t2 − 7/6 t3 4d) 1+e−2 s −2 e−s
(1−e−2 s )s
2o) H (t − 5) t 1
5c) s
(e −1)s
2p) H (t − 6) (5 cos (t − 6) + sin (t − 6))
6a) 1 + H (t − 1) + H (t − 2) + H (t − 3) + H (t − 4) + · · ·
6b) 1 − 1/2 H (t − 1) + 3/8 H (t − 2) − 5 H 16
(t−3)
+ 35 H128
(t−4)
+ ···
3 4
2 t 5t
6c) 1 + t/2 − 1/16 t + 96 − 3072 + + · · ·
Section 6.3, pg 158
1a) −1 + (s − 4) Y 2a) 1/2 et − 1/2 cos (t) − 1/2 sin (t)
1b) 4 + (2 s + 7) Y 2b) 1/2 t sin (t)
1c) s2 Y + 5 Y +
2 s + 1 2c) cos t(sin t − 2 cos t) + 1 + e−t
1d) s2 + 3 s −
2 Y − s 2d) −1/2 sin (t) + 1/2 sinh (t)
1e) 4 s2 + 2 s Y + 8 s − 2 2e) 1/2 t2 + cos (t) − 1
Section 6.4, pg 163
1a) 1 + e−t/2 2b) −1 + cos (2 t) + 2 t2
1b) −1/2 e−t + e−t/3 2c) et − sin (t) + cos (t) − 2
1c) 1/3 e−2 t − 1/3 et 2d) 1/2 t2
1d) 2te3 t 2e) 1 − 1/2 et cos (2 t) − 1/2 e−t
Rt
1e) 4 − 5 et/5 3a) 0 ln (1 + 3 τ ) e−3 t+3 τ dτ
1f) −2 sin (t/2) − cos (t/2) R t√
t
3b) 13 0 1 + τ sin (3 t − 3 τ ) dτ
1g) −e cos (t) R t −2 t+2 τ R t t/2−τ /2
1h) −3 e−t sin (2 t) 3c) − 15 0 e 1+τ dτ + 15 0 e 1+τ dτ
Rt
2a) 4 et − e−2 t − 6 t − 3 3d) 12 0 sin(1+τ 2 )e−t+τ sin (2t − 2τ ) dτ
Rt
4a) 0 ln (1 + 3 τ ) e−3 t+3 τ dτ + e−3 t
R t√
4b) cos (3 t) + 31 0 1 + τ sin (3 t − 3 τ ) dτ
R t −2 t+2 τ R t t/2−τ /2
4c) 8/5 e−2 t + 2/5 et/2 − 15 0 e 1+τ dτ + 51 0 e 1+τ dτ
Rt
4d) e−t sin (2 t) + 12 0 sin(1 + τ 2 )e−t+τ sin (2t − 2τ ) dτ
Section 6.5, pg 169
1a) e−4 t + 3/4 H (t − 1) 1 − e−4 t+4

1b) −1 + et/2−2 H (4 − t) − et/2−2

1c) 2 H (t − 3) e−t+3 − e−t

1d) −1/2 H (t − 2) + 1/2 (1 − H (2 − t)) e4 t−8
+e
4 t−4
(H (1 − t) − 1)
−2 t+10 3 t−15
1e) 1/10 H (t − 5) −5 + 3 e + 2e
2 2
1f) 3/2 H (t − 4) (sin (t − 4))
− 3/2 H (t − 2) (sin (t − 2))
1g) 3/4 H (t − 1) −1 + e4 t−4
1h) −2 H (t − 2) sin (t − 2) + H (t − 3) sin (t − 3)
Section 6.6, pg 173
18 e−3 t 5/2 e2 t + 3/2 e4 t
" # " #
5 + 2/5 e2 t
1a) 1c)
−6/5 e−3 t + 1/5 e2 t −5/2 e2 t + 3/2 e4 t
" 13 5t
#
5 + 7/5 e
t/2
" #
7/4 e + 9/4 e −t/2
1b) 1d)
21 e5 t
7/2 et/2 − 9/2 e−t/2 − 26
5 + 5
232 Appendix B. Answers

4 e2 t
" # " #
s 1/4
1
1e) 3b) s2 −1/4
−e2 t − 4 e2 t t 1 s
" 1 t
− 2 e sin (2 t) + 4 et cos (2 t)
# " #
s−2 0
1
1f) 3c) (s−2)2
−et cos t
" (2 t) − 8 e# sin (2 t) "−1 s − 2 #
s 6 s − 1 −4
1 1
3a) s2 +s−6 3d) s2 −2 s+5
1 s+1 1 s−1

Chapter 7
Section 7.2, pg 179
e2 x e−2 x
1a) u (x) = − e−4 −e4 + e−4 −e4
1b) u (x) = − sin(x)
sin(1)
√
e−x/2 sin(1/2 3x)
1c) u (x) = e−1/2 sin(1/2
√
3)
e−x (−1+e2 ) ex (−1+e−2 )
1d) u (x) = −5 e−2 −e2 + 5 e−2 −e2 − 5
e−x

−1
1e) u (x) = 1/2 x2 + 1/2 e−1 −1 − x − 1/2 e−1 − 1
un = bn sin π2 (2n − 1)x , with λn = −[ π2 (2n 2
 
3a)
− 1)]
nπ
3b) u0 = b0 , with λ0 = 0; and un = bn cos 4 x , with λn = −( nπ 4 )
2
p
3c) u = be−λx/2 sin(πx/4), with λ = ± 4 − (π/2)2
3d) un = bn e−x/2 sin(nπx), with λn = − 14 − (nπ)2
3e) u0 = b0 , λ0 = 0; un = an sin(2πnx) + bn cos(2πnx), with λn = 4π 2 n2
Section 7.3, pg 186
2
1a) −4 e−75 π t sin (5 π x)
2
1b) 6 e−363 π t sin (11 π x)
2 2 2
1c) e−3 π t sin (π x) + 8 e−48 π t sin (4 π x) − 10 e−147 π t sin (7 π x)
2 2 2
1d) −e−27 π t sin (3 π x) + 7 e−192 π t sin (8 π x) + 2 e−675 π t sin (15 π x)
2 2
1e) 2 e−27 π t sin (3 π x) + 2 e−3 π t sin (π x)
n −n2 π 2 t
2a) n=1 −2 (−1+(−1) )e n π sin(1/2 n π x)
P∞
n −n2 π 2 t
2b) n=1 −4 (−1+2 (−1) )e n π sin(1/2 n π x)
P∞
−π 2 t n −n2 π 2 t
2c) −4/3 e sin(1/2 π x)
+ n=3 −2 n (−1+(−1) )e sin(1/2 n π x)
P∞
 π  π (n2 −4)

2d) n=1 2 cos(1/2 n π) 2 2

− 2 n1π e−n π t sin (1/2 n π x)
P∞
nπ
P∞  n

2e) n=1 −4 (−1) +2n πcos(1/6 n π) + 2 n1π e−n π t sin (1/2 n π x)
2 2

n −n2 π 2 t
3) n=1 30 (−1) e sin(1/3 n π x)
P∞
nπ
P∞ (2 cos(1/4 n π)−2 cos(3/4 n π))e−5/2 n2 π2 t sin(1/2 n π x)
4) n=1 nπ
2
5a) n=1 bn e−kn t sin(kn x), kn = π(2n − 1)/2
P∞
2
5b) n=1 bn e−4kn t cos(kn x), kn = π(2n − 1)/2
P∞
2 2
5c) n=1 bn e−kn (t+t /2) sin(kn x), kn = nπ
P∞
2
5d) n=1 bn e−kn t+e sin(kn x), kn = nπ
P∞ −t

2 2
6d) u = −24e(−k3 t+e −1) sin(k3 x) − 12e(−k15 t+e −1) sin(k15 x)
−t −t

Section 7.4, pg 198

n
4a) n=1 −4 (−1) sin(1/2 n π x)
P∞
nπ
n 2
P∞ n π ((−1) e −1) sin(1/2 n π x)
4b) n=1 −2 n2 π 2 +4
 233

∞
X n sin (1/2 n π x)
4c) 4
n=1
π (n2 − 4)
n odd
n
 
6 (−1) −8 cos(1/4 n π)
+ 2 n1π sin (1/2 n π x)
P∞
4d) n=1 nπ
P∞  
4e) n=1 −2 n π cos(1/2nn2π)−4 π2
sin(1/2 n π)
+ 4 n1π sin (1/2 n π x)
P∞  n

4h) n=1 −6 (−1) −2 cos(1/6 nπ
n π)+6 cos(2/3 n π)
+ 2 1
n π sin (1/2 n π x)
P∞ ((−1)n −1) cos(1/2 n π x)
5a) 1 + n=1 4 n2 π 2
2
P∞ ((−1)n e2 −1) cos(1/2 n π x)
5b) −1/2 + 1/2 e + n=1 4 n2 π 2 +4
5c) cos(πx)
5d) −2 + n=1 8 sin(1/4 n π)ncos(1/2 n π x)
P∞
π 
5e) 3/4 + n=1 2 n π sin(1/2 nnπ)−4 cos(1/2 n π) 1
P∞
2 π2 + 4 n2 π 2 cos (1/2 n π x)
P∞ (2 sin(1/6 n π)−6 sin(2/3 n π)) cos(1/2 n π x)
5h) 7/6 + n=1 nπ
n
8a) 1/3 + n=1 4 (−1) ncos(n π x)
P∞
2 π2
n
9a) n=1 −2 (−1) nsin(n π x)
P∞
π
Section 7.5, pg 204
1a) cos(12πt) sin(3πx)
1
1b) − 16π sin(32πt) sin(8πx)
1c) − cos (4 π t) sin (π x) + 4 cos (12 π t) sin (3 π x) − 3 sin(20 π20t)πsin(5 π x)
1d) 5 cos (28 π t) sin (7 π x) + sin(32 π 16π t) sin(8 π x)
+ sin(48 π t)16π
sin(12 π x)

sin(32 π t) sin(8 π x)
1e) cos (12 π t) sin (3 π x) + cos (4 π t) sin (π x) − 16π
1f) 3Pcos (8 π t) sin (2 π x) − sin(36 π 24π t) sin(9 π x)

+ 3 sin(20 π40t)πsin(5 π x)
∞
2a) n=1 an cos(kn t) + bn sin(kn t) sin(k
n x), kn = (2n − 1)π/2
P∞
2b) n=1 an cos(2kn t) + bn sin(2kn t) cos(k
n x), kn = (2n − 1)π/2

P∞
2c) a + bt + n=1 an cos(nπt) + bn sin(nπt) An cos(2nπx) + Bn sin(2nπx)


√
2d) e−t/2 n=1 an cos(ωn t) + bn sin(ωn t) sin(nπx) , ωn = 4n2 π 2 − 1/2
P ∞
n
3) n=1 −2 (−1+(−1) ) sin(2 n π t) sin(n π x)
P∞
n4 π 4
Section 7.6, pg 207
n −4 n2 π 2 t
1a) 1 − 2 x + n=1 −2 (1+(−1) )e n π sin(n π x)
P∞
n −4 n2 π 2 t
1b) 2 − 7 x + n=1 −14 (−1) e n π sin(n π x)
P∞
−4 n2 π 2 t
1c) −4 + 5 x + n=1 8 e sin(n π x)
P∞
nπ
2a) 1 − x
2b) −7 + 2x
2c) −1 + 3x
2d) Aex + Be−x , where A = (2 − e−3 )/(e3 − e−3 ), B = (e3 − 2)/(e3 − e−3 )
2e) A + Bex , where A = (1 + e2 )/(1 − e2 ), B = 2/(e2 − 1)
n −n π (t+1/2 t )
2 2 2

4) −1 + 3 x + n=1 2 (1+2 (−1) )e sin(n π x)

P∞
nπ
5) 1 − 2 x − 7 cos (9 π t) sin (3 π x)
Section 7.7, pg 211
(4 k5 cos(t)+4 sin(t)−4 k5 e−k5 t ) sin(5 π x)
1a) k5 2 +1
, where k5 = 100π 2
e−k3 t (−1+ek3 t−2 t ) sin(3 π x)
1b) − k3 −2 , where k3 = 36π 2
2 2
 
P∞ (−1+(−1)n ) e−4 n π t −1 sin(n π x)
1c) n=1 −1/2 n3 π 3
234 Appendix B. Answers

n −(5+1/4 π n )t
2

3d) n=1 −4 (−1) e sin(1/2 n π x)

P∞
nπ
Section 7.8, pg 220
1a) 5 sinh(2πy) sin(2πx)/ sinh(4π)
1b) −3 sinh(12πy) sin(12πx)/ sinh(24π)
1c) sinh(πy) sin(πx)/ sinh(2π) − 7 sinh(8πy) sin(8πx)/ sinh(16π)
1d) −3 sinh(4πy) sin(4πx)
sinh(8π) − sinh(7πy) sin(7πx)
sinh(14π) + 6 sinh(20πy) sin(20πx)
sinh(40π)
2a) 21 r3 cos(3θ)
2b) 1 − 3(r/2)15 sin(15θ)
2c) (r/2) sin(θ) + 3(r/2)5 cos(5θ))
2d) 4 − 2(r/2)5 sin(5θ) − 4(r/2)9 sin(9θ) + 8(r/2)14 cos(14θ)
3d) 7 sinh(3πx) sin(3πy)/ sinh(3π)
3e) −2 sinh(2πx) sin(2πy)/ sinh(2π) + 8 sinh(7πx) sin(7πy)/ sinh(7π)
4b) n=1 cn r2n sin(2nθ)
P∞
4c) −3(r/2)4 sin(4θ)
4d) 9(r/2)2 sin(2θ) − 5(r/2)14 sin(14θ)
 Bibliography

G. Bonani, S. Ivy, W. Wolfli, M. Broshi, I. Carmi, and J. Strugnell.

Radiocarbon dating of fourteen Dead Sea scrolls. Radiocarbon, 34(3):
843–849, 1992. doi: 10.1017/S0033822200064158.
A. Coddington and R. Carlson. Linear Ordinary Differential Equations.
Society for Industrial and Applied Mathematics, 1997. ISBN 978-0-
898713-88-6.
A. M. Cohen. Numerical Methods for Laplace Transform Inversion. Nu-
merical Methods and Algorithms. Springer, 2007. ISBN 978-0-387-
28261-9.
B. Davies. Integral Transforms and Their Applications. Texts in Applied
Mathematics. Springer-Verlag New York, 3d edition, 2002. ISBN 978-
0-387-95314-4.
D. Eigler. Quantum Corral. Website, 2020. http://www.nisenet.org/
catalog/scientific-image-quantum-corral-top-view.

R. Engbert and F. Drepper. Chance and chaos in population biology -

models of recurrent epidemics and food chain dynamics. Chaos, Soli-
tons, and Fractals, 4(7):1147–1169, 1994. doi: 10.1016/0960-0779(94)
90028-0.
B. Friedland. Control System Design: An Introduction to State-Space
Methods. Dover Publications, Inc., New York, NY, USA, 2005. ISBN
978-0486442785.
K. E. Gustafson. Introduction to Partial Differential Equations and
Hilbert Space Methods. Dover Publications, 3d edition, 1999. ISBN
9780486140872.
M. H. Holmes. Introduction to the Foundations of Applied Mathematics.
Texts in Applied Mathematics. Springer, 2nd edition, 2019. ISBN 978-
0-387-87749-5.
R. F. Hoskins. Delta Functions: Introduction to Generalised Functions.
Woodhead Publishing, 2nd edition, 2009. ISBN 978-1-904275-39-8.

235
236 BIBLIOGRAPHY

A. J. Jerri. The Gibbs Phenomenon in Fourier Analysis, Splines and

Wavelet Approximations. Mathematics and Its Applications. Springer,
1998. ISBN 978-0-7923-5109-2.

NASA. Drag of a sphere. Website, 2020. https://www.grc.nasa.gov/WWW/

K-12/airplane/dragsphere.html.

F. Oberhettinger and L. Badii. Tables of Laplace Transforms. Springer-

Verlag Berlin Heidelberg, 1973. ISBN 978-3-540-06350-6.

A. E. Parker. Who solved the Bernoulli differential equation and how

did they do it? College Mathematics Journal, 44(2):89–97, 2013. doi:
10.4169/college.math.j.44.2.089.

L. Perko. Differential Equations and Dynamical Systems. Texts in Applied

Mathematics. Springer, 2001. ISBN 978-0-387-95116-4.

F. W. Roos and W. W. Willmarth. Some experimental results on sphere

and disk drag. AIAA Journal, 9(2):285–291, 1971. doi: 10.2514/3.6164.

I. Stakgold. Boundary Value Problems of Mathematical Physics (Vol.

1). Society for Industrial and Applied Mathematics, Philadelphia, PA,
USA, 2000. ISBN 978-0-89871-456-2.

W. A. Strauss. Partial Differential Equations: An Introduction. Wiley,

2nd edition, 2007. ISBN 978-0-470-05456-7.

A. M. Stuart and A. R. Humphries. Dynamical Systems and Numeri-

cal Analysis. Cambridge Monographs on Applied and Computational
Mathematics. Cambridge University Press, 1998. ISBN 978-0-521-
64563-8.

G. P. Tolstov and R. A. Silverman. Fourier Series. Dover Publications,

1976. ISBN 978-0486633176.

D. V. Widder. Laplace Transform (PMS-6). Princeton University Press,

1941. ISBN 9780691627755.
 Index

H(·), 153 periodic, 180, 217 first-order linear, 14

W (y1 , y2 ), 43 wave equation, 199 first-order system, 79
L(·), 145 boundary value problem, 176 homogeneous, 3
δ(t), 165 Brownian motion, 176 independent variable, 3
det(·), 86 buoyant force, 30 linear, 3
∀, 43, 85 BVP, 176, 200 order, 3
I, 85 second-order linear, 41
∇2 , 212 catenary, 13 diffusion equation, 175
Im(·), 47 center, 108 inhomogeneous, 208
Re(·), 47 central force field, 139 inhomogeneous boundary
g, 23, 64, 72 characteristic equation, 45, conditions, 205
i, 46 48, 176 separation of variables, 180
eigenvalues, 86 steady state, 205
Abel’s formula, 44 compatibility conditions discontinuous forcing
acceleration, 2, 23, 41, 79, wave equation, 207 function, 164
176 complex conjugates, 90 distribution, 168
advection equation, 175 complex number, 47 drag force, 23
Airy’s equation, 163 imaginary part, 47 on sphere, 31
amplitude, 65 real part, 47 driving frequency, 70
Archimedes’ principle, 30, 73 convolution theorem, 158 Dulong-Petit law of cooling,
associated homogeneous cosine series, 193 27
equation convergence theorem, 194
first-order equation, 17 differentiability, 197 eigenvalue, 85
matrix equation, 18 critical point, 38 eigenvalue problem, 86
second-order equation, 42, BVP, 178
52, 57 damped eigenvector, 85
asymptotically stable, 35 critically damped, 75 independent, 88
linear system, 107 over-damped, 75 electrostatic force, 140
nonlinear system, 117–119 damped wave equation, 204 epidemic equilibrium, 125
autonomous equation, 35, damping epidemics, 80
113 critically damped, 68 equilibrium point, 38
over-damped, 67 Euler equation, 76
balance law, 176 under-damped, 68 Euler’s formula, 46
bang-bang wave, 156 weakly damped, 69, 71 existence and uniqueness
baseball, 31 damping constant, 67 theorem, 42
beam equation, 6 Dead Sea scrolls, 29 exponential order, 147
Bernoulli equation, 19 defective matrix, 93
Bessel equation, 63 delta function, 165 Fejér summation, 193
Beverton-Holt model, 31 determinant, 43, 86, 127 floor function, 157
boundary conditions, 177 differential equation flutter, 70
diffusion equation, 181 dependent variable, 2 forcing

237
238 Index

periodic, 75 Jacobian matrix, 119, 121 neutrally stable, 36, 108,

forcing amplitude, 70 joke model, 128 118, 119
forcing function, 41 jump discontinuity, 146, 153, Newton’s law of cooling, 26
oscillator, 64 188 Newton’s second law, 2, 23,
Fourier sine series, 184, 188 41, 64, 79, 129, 131,
Kermack-McKendrick 139, 176
Galerkin method, 211 model, 80
general solution ODE, 3
diffusion equation, 183 Laplace transform, 145 one-sided stability, 40
first-order equation, 17 convolution, 158 oscillator
linear system, 84, 91 impulse forcing, 165 Duffing, 116, 137
second-order equation, 42, inverse, 150 Morse, 138
57 of derivative, 157 simple harmonic, 65, 129
wave equation, 201, 215 periodic function, 156 Toda, 116
Gibbs phenomenon, 193 solving differential Van der Pol, 116
gravitational acceleration equations, 159
constant, 23, 64, 111 table, 151 partial differentiation
gravitational force, 23, 64, Laplace’s equation, 212 notation, 175
140 periodic boundary partial fractions, 160, 169
conditions, 217 particular solution, 17
libration, 134 non-uniqueness, 52
half-life, 28
linear approximation, 119 second-order equation, 52
half-plane of convergence,
linear operator, 149 PDE, 3
146
linear system pendulum, 6, 111, 132
Hamiltonian, 131, 137, 139
general solution, 84 period, 136
Hamiltonian system, 139
homogeneous, 82, 170 periodic forcing, 75
Heaviside Expansion
inhomogeneous, 170 periodic orbit, 141
Theorem, 155
linearized stability theorem, periodic solution, 65, 129,
Heaviside step function, 153
121 137
derivative, 168
linearly independent phase, 65
Hilbert space, 197
equating coefficients, 56 phase plane, 99
homicide victim, 32
functions, 43 phase portrait, 99, 121, 122
homogeneous, 3
vector functions, 84 table, 100
Hooke’s law, 64
vectors, 86, 88 piecewise continuous, 147,
Wronskian, 43 188
identity matrix, 85 logistic equation, 25 predator-prey equations, 116
impulse, 166 principle of superposition, 5
impulse forcing, 165 mass-spring-dashpot, 6, 64 linear system, 83
indeterminate steady state, matrix PDEs, 183
121 defective, 90
inhomogeneous, 3 identity, 85 radioactive decay, 1, 6, 28
inhomogeneous boundary non-invertible, 86 Rayleigh quotient, 179, 180
conditions, 205 singular, 86 reduction of order, 45, 76
initial condition, 2 Maxwell viscoelastic resonance, 70
diffusion equation, 181 material, 19
separation of variables, 183 measles, 112, 126 saddle, 108, 122
initial conditions method of undetermined saddle point, 123, 125
second-order equation, 42 coefficients, 53 sawtooth wave, 156
wave equation, 199 first-order equation, 18, 59 Schrödinger’s equation, 6
initial value problem, 2 Michaelis-Menten equations, separable equation, 7
second-order equation, 42 6, 116 separation of variables
integral curves, 99 mixing problems, 20 for ODE, 8
integrating factor, 15, 209 for PDEs, 181, 200, 214,
isolated steady states, 115 natural frequency, 65, 204 218
IVP, 2 natural mode, 204 non-uniqueness, 11
Index 239

separation constant, 182 nonlinear system, 121 unstable, 35

separatrix, 135 single equation, 35 linear system, 107
simple harmonic motion, 65 standing wave, 204 nonlinear system, 120
sine series, 184 steady state, 35
convergence theorem, 189 linear system, 107
nonlinear system, 114 variation of parameters, 20,
differentiability, 196
PDE, 205 60
sink, 108, 122, 124
velocity, 2, 23, 79
SIR model, 80, 112 Temple Scroll, 29 angular, 111
Somigliana equation, 23, 72 term-by-term differentiation,
source, 108, 122 196
spiral sink, 108, 122, 123 terminal velocity, 24, 30, 31 wave equation, 175
spiral source, 108, 122 trace, 110, 127 compatible boundary
spring constant, 64 transfer function, 161 conditions, 207
square wave, 156 systems, 173 damped, 204
stability theorem triangle wave, 156 weight, 23, 72, 73
linear system, 108 turkey, 32 Wronskian, 43, 61, 85